Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[v0.292.2] - 2026-06-09 - Infrastructure Network Enhancements

Changed

🚀 Improved Backend Network Connectivity: The core backend services are now connected to a dedicated egress network across all deployment configurations, enhancing support for secure and managed outbound connections.

[v0.292.0] - 2026-06-08 - Streamlined Python Distribution: PyPI Launch and Enhanced Documentation

Added

🚀 PyPI Publishing Workflow: Introduced a new automated GitHub Actions workflow for building and publishing all public Python SDK packages to PyPI, enabling wider distribution and easier installation.
🔐 Distribution Artifact Validation: Added a new Python script (check_dist.py) to enforce namespace integrity, prevent secret leaks, and flag unexpectedly large files within PyPI packages, significantly enhancing release security and quality.
📦 swiss-ai-hub Meta-Package: A new meta-package, swiss-ai-hub, has been added to simplify the installation of the entire Swiss AI Hub Python SDK with a single command.
📄 Package License Files: Explicit LICENSE files have been added to each Python package (agent, api, bot, core, meta, pipeline, process), ensuring clear license compliance for distributed artifacts (Apache-2.0 for most, AGPL-3.0-or-later for backup).

Changed

📝 Comprehensive Package READMEs: The README.md files for all Python SDK packages have been significantly expanded and improved to provide detailed purpose, installation, usage, development, and production guidance, tailored for PyPI consumers.
⚙️ Standardized PyPI Metadata: All pyproject.toml files across the Python SDK packages have been enhanced with detailed descriptions, relevant PyPI classifiers, and comprehensive project URLs (Homepage, Repository, Documentation, Issues) for improved discoverability and information.
🔗 Pinned Internal Dependencies: Python pyproject.toml files now explicitly pin inter-package dependencies (e.g., swiss-ai-hub-agent depending on swiss-ai-hub-core) to their exact version, ensuring consistent and predictable builds.
🛠️ Build System Configuration: Adjusted uv_build settings in pyproject.toml for all Python packages to correctly handle PEP 420 native namespaces during artifact creation, ensuring proper module loading.
🌐 Public Image URLs in README: All image references in the root README.md have been updated to use absolute GitHub raw URLs, guaranteeing they display correctly on external platforms like PyPI.
🔒 NPM Publish Workflow Environment: The publish-npm.yml GitHub Actions workflow now uses an explicit npm environment for enhanced security context during package publication.
📖 Environment Variables Documentation: The formatting and readability of the environment variables reference documentation have been updated for better clarity.
⬆️ Core Dependency Updates: Several key dependencies in swiss-ai-hub-core have been upgraded, including mem0ai, langchain-neo4j, neo4j, rank-bm25, and starlette (with a security floor to 1.0.1), enhancing functionality and addressing known vulnerabilities.
🐍 jambo Internal Dependency: The jambo dependency in swiss-ai-hub-api has been migrated to swiss-ai-hub-jambo, reflecting an internal packaging change for consistent namespace usage.
⚡️ Version Bump Script: The Makefile's version-bump target now correctly updates inter-package dependencies to ensure all internal swiss-ai-hub-* packages reference the new version.

Refactor

🧹 Form Default Seeding Logic: The web UI's seedFormDefaults function has been refactored by extracting dedicated helper functions for handling group and repeater elements, improving code clarity and maintainability.

[v0.291.13] - 2026-06-08 - Enhanced OpenAI Service Iframe Permissions

Changed

⚡️ Expanded OpenAI service iframe capabilities: The embedded OpenAI service now supports additional browser features such as clipboard access, camera, screen capture, fullscreen mode, geolocation, and autoplay, alongside refined microphone access.

[v0.291.12] - 2026-06-08 - Refined Skill Permissions for Enhanced Control

Changed

⚡️ Refined Skill Permissions: The broad mcp__* skill wildcard has been replaced with a specific, enumerated list of allowed mcp skills, enhancing control and clarity over which tools are accessible.

[v0.291.11] - 2026-06-08 - Enhanced Changelog Documentation

Changed

⚡️ Improved Changelog Documentation Handling: The documentation build script now correctly processes changelog content by wrapping its body in a v-pre container. This change prevents build errors caused by special syntax (e.g., GitHub Actions expressions) within release notes, ensuring that content is rendered verbatim on the documentation site while preserving the functionality of dynamic elements like copy/download buttons.

[v0.291.10] - 2026-06-08 - Enhanced Security and Code Quality

Added

✨ Introduced CodeQL Analysis workflow: A new GitHub Actions workflow has been added to perform automated static code analysis, enhancing the security and overall code quality of the Python and JavaScript/TypeScript codebases.

[v0.291.9] - 2026-06-08 - Simplified Access Control and Enhanced Identity Provider Setup

Added

📄 New Identity Provider Setup Documentation: Introduced comprehensive guides for connecting external identity providers to the Swiss AI Hub via Keycloak, starting with detailed instructions for Microsoft Entra ID (Azure AD).
🔐 Azure App Registration Guide: Provided step-by-step documentation for configuring Azure App Registrations, including redirect URIs, required API permissions (openid, email, profile), and client secret management.
👥 Azure User and Role Management Guide: Added clear instructions on defining and assigning AIHubAccess (mandatory for platform login) and AIHubSysAdmin app roles within Azure Entra ID to control user access.

Changed

🔑 Updated Superuser Role Configuration: Modified default superuser roles (SUPERUSER_ROLES_JSON) to explicitly include AIHubAccess as the base role for platform usage, aligning with the refined access control model.
📖 Enhanced Authentication & Authorization Documentation: Significantly updated the security documentation to clarify the distinction between Keycloak Realm roles (AIHubAccess, AIHubSysAdmin) and platform-managed tenant-scoped roles, providing a clearer understanding of the permission hierarchy.

Removed

🗑️ Streamlined Keycloak Realm Roles: Deprecated and removed AIHubAdmin, AIHubUser, and AIHubDeveloper from the Keycloak realm configuration. These roles are now intended to be managed as tenant-specific roles within the platform's internal access management system.
🧹 Simplified Identity Provider Mappers: Removed corresponding identity provider mappers for the deprecated AIHubAdmin, AIHubUser, and AIHubDeveloper roles, streamlining the IdP integration process.

Added

✨ Seamless Initial Navigation: Introduced a new, dedicated middleware for the home route (/) to intelligently determine and redirect users to their appropriate tenant, tenant selection, or previously visited page immediately after login, ensuring a smoother start to their session.
🚀 Global Loading Overlay: Implemented a full-screen loading spinner that instantly appears when the initial navigation logic is resolving, eliminating content flicker and providing clear visual feedback during critical redirects.
⚙️ Centralized Home Resolution State: Added a shared state management system and client-side plugin to accurately manage the global loading indicator, ensuring it correctly displays during initial route resolution and dismisses on navigation completion or errors.

Changed

🛡️ Robust Authentication Handling: Enhanced the authentication flow to automatically invalidate and remove problematic user sessions when silent sign-in or token renewal encounters errors, improving stability and preventing repeated authentication failures.

Refactor

🧹 Refined Initial Login Flow: Streamlined the application's startup by relocating the logic for checking and renewing user sessions from the OIDC client plugin to a global authentication middleware, allowing for more efficient and non-blocking initialization.
⬆️ Modernized Homepage Implementation: Replaced the client-side onMounted logic for tenant fetching and redirection on the homepage with a server-side (or early client-side) middleware, significantly improving the responsiveness and consistency of the initial user experience.

[v0.291.7] - 2026-06-03 - Enhanced Model Cost Reporting and API Robustness

Added

✨ Introduced Granular Search Context Cost Reporting: Added a new data transfer object (SearchContextCostPerQueryDTO) to capture and display search context costs broken down by low, medium, and high context sizes, providing more detailed pricing information.
🦾 Enhanced API Robustness for Model Information: Implemented a new data validation mechanism that gracefully handles unexpected data shapes in external model metadata feeds by logging warnings and dropping malformed fields, preventing parsing failures for the entire model information.

Changed

📊 Updated Model Details Panel: The Model Details Panel now presents search context costs with a more granular breakdown (low, medium, and high context sizes), offering clearer insights into model pricing.
🌍 Localized Search Context Cost Labels: Updated internationalization files to include specific labels for low, medium, and high search context costs in various languages (English, German, French, Italian).
🔄 Revised Model Information Schema: Modified the ModelInfoDTO to incorporate the new granular SearchContextCostPerQueryDTO for search context costs and made the mode field optional, reflecting more flexible model metadata.

Changed

🚀 Enhanced LLM Guard Performance: Switched agent description and few-shot guards to use asynchronous structured prediction, improving responsiveness and concurrency with Large Language Models.
⚡️ Improved Concurrency for Reranking Nodes: Blocking reranking operations are now offloaded to a separate thread using asyncio.to_thread, preventing event loop starvation and boosting application performance.
📄 Refined Error Logging: Updated MineruLoader and Redis request hooks to use logger.exception, ensuring comprehensive exception details are automatically included in logs for better debugging.
🛠️ Corrected OpenTelemetry HTTPX Instrumentation: Ensured asynchronous HTTPX request and response hooks are properly defined as coroutine functions, enabling full observability for async HTTP communications.

Refactor

🧹 Optimized Dispatcher Event Handling: The _build_event_kwargs method in agent and process dispatchers, as well as the base dispatcher, has been made synchronous, streamlining event argument construction.
🔄 Standardized Access Rule Prefixes: Consolidated aihub.admin and aihub.user access rule prefixes into dedicated constants, enhancing maintainability and clarity across authentication logic.
🔑 Centralized Authentication Error Messages: Moved common error messages in the authentication handler to constants, improving consistency and simplifying future updates.
☁️ Standardized S3 Service Error Messages: Consolidated error messages for S3 anonymous file access service operations into constants, improving consistency and maintainability.
⚙️ Refined SonarQube Configuration: Added specific ignore rules to sonar-project.properties to suppress python:S1192 (duplicated string literals) warnings for known-good patterns, such as lazy imports and MongoDB query syntax, reducing false positives in code quality scans.
📝 Improved RPC Success Attribute Handling: Standardized the rpc.success attribute for NATS requester tracing using a dedicated constant.
📚 Improved Test Mock Clarity: Added noqa comments to asynchronous mock methods in authentication utilities and Milvus vector store tests to explicitly state why they remain async for testing purposes, improving code clarity for linters.
🧹 Minor Python Idiom Improvements: Updated set creation from set(generator_expression) to {comprehension} in base_event and dispatchable_workflow for more idiomatic Python.

[v0.291.5] - 2026-06-02 - Enhanced FormKit Data Handling for Stability

Changed

🔄 Refined getNestedValue utility: The getNestedValue function has been refactored to operate as a pure read-only function, specifically for retrieving nested arrays. This change prevents unintended mutations of form data during render-time and mitigates potential recursive render loops in reactive contexts. It now strictly returns an array or an empty array if the specified path does not lead to an array.

Fixed

🛠️ Improved Repeater Initialization: FormKit repeater fields now correctly materialize as an empty array during form loading if their default value is undefined. This proactive initialization ensures a consistent array structure for repeaters, preventing unexpected behavior and improving stability in reactive render contexts.

[v0.291.4] - 2026-06-02 - Enhanced Prompt Reliability for Image Rendering

Fixed

🐛 Image Rendering in Prompts: Resolved a Jinja SecurityError that occurred when attempting to render image URLs (pydantic.AnyUrl) within RAG agent and guard context prompts. The rendering now correctly uses the | string filter to safely convert URLs, ensuring images are displayed as intended across all supported languages.

[v0.291.3] - 2026-06-02 - Stricter OAuth Role Defaults

Changed

🔑 Updated OAuth Allowed Roles Default: The default OAUTH_ALLOWED_ROLES configuration for Open WebUI has been updated across all deployment configurations. It now explicitly requires AIHubAccess and any configured admin roles (${OAUTH_ADMIN_ROLES_OPENWEBUI}) instead of a wildcard (*), enhancing access control and improving the security posture for OAuth integrations.

[v0.291.2] - 2026-06-02 - GitHub Actions Reliability Improvement

Fixed

🐛 Corrected GitHub Actions if condition parsing: Ensured proper evaluation of the claude-code-review workflow's trigger conditions by explicitly wrapping the complex expression in ${{ ... }}.

[v0.291.1] - 2026-06-02 - Improved Code Quality Checks with SonarCloud Action v6

Changed

🚀 Updated SonarCloud Scan Action: The GitHub Action for running SonarCloud code quality scans has been upgraded to v6.0.0, enhancing reliability and ensuring compatibility with the latest SonarCloud features.
📄 Documentation Refresh: Internal documentation (CLAUDE.md) was updated to reflect the new version of the SonarCloud scan action used in CI/CD workflows.

[v0.291.0] - 2026-06-02 - 🚀 Web Package Stability, Deployment, and Publishing Overhaul

Added

🚀 Automated npm publishing workflow: Introduced a dedicated GitHub Actions workflow (publish-npm.yml) to streamline and automate the publishing of the @swiss-ai-hub/web package to npm, utilizing OIDC trusted publishing for enhanced security.
🧪 Web package build and pack verification: Added a new GitHub Actions workflow (verify-web-package.yml) to automatically build and validate the packaging of @swiss-ai-hub/web on every pull request, ensuring package integrity and publishability before merge.

Changed

⚙️ Streamlined CI/CD branch triggering: Modified several core CI/CD workflows (analyze-test-pr, lint-pr, semantic-pr, test-backup-e2e) to exclusively trigger on the main branch, simplifying branching strategy and pipeline execution.
📄 Enhanced web package installation and dependency management documentation: Significantly updated the packages/web/README.md with detailed guidance on required dependency overrides (specifically for vue and primevue) to prevent common runtime issues caused by multiple framework instances.
🔄 Redesigned runtime configuration for static builds: Introduced a new, more robust runtime configuration mechanism for the @swiss-ai-hub/web layer, moving away from NUXT_PUBLIC_* environment variables to a dynamic /config.js file generated at container startup, offering greater deployment flexibility for static single-page applications.
🐳 Updated Dockerfile and deployment examples: Provided comprehensive and up-to-date Dockerfile and nuxt.config.ts examples in the packages/web/README.md that reflect the new best practices for building, serving, and configuring the static web package in production environments.
✨ Integrated FormKit PrimeVue: Added @sfxcode/formkit-primevue as a new dependency within the packages/web package, enhancing its form building capabilities with PrimeVue components.
⬆️ Updated PrimeVue peer dependency: Bumped the recommended primevue peer dependency version to 4.5.5 to ensure compatibility and leverage the latest updates for the UI component library.

[v0.290.13] - 2026-06-02 - OAuth Default Role Access Update

Added

✨ Introduced OAUTH_ALLOWED_ROLES configuration: Added the OAUTH_ALLOWED_ROLES environment variable, set to * by default across Docker Compose configurations, to explicitly allow all roles from the OAuth provider, simplifying initial setup for role-based access.

Changed

🔗 Updated External README Links: Transformed relative paths for licenses and contributing guidelines in the main README.md to absolute GitHub URLs, ensuring links function correctly across all platforms.
📄 Minor Documentation Formatting: Improved the display of the Dagster UI URL in the packages/backup README for better readability.

Refactor

🧹 Standardized Internal Documentation Paths: Performed a comprehensive review and adjustment of numerous relative links throughout the German and English documentation, enhancing navigation and consistency within the platform guides and SDK.
📚 Streamlined Authentication Documentation: Removed outdated architectural decision record (ADR) links from authentication-related documentation, simplifying the content and focusing on current implementation details.

[v0.290.11] - 2026-06-01 - Enhanced Documentation and Licensing Clarity

Added

📄 New API Tokens Documentation: A dedicated guide has been added explaining how to generate, use, and revoke personal API tokens for REST API authentication, including steps within the Swagger UI.
✨ Comprehensive SDK Licensing Rationale: New documentation now details the platform's mixed-license model, clarifying the rationale behind using Apache 2.0 for the backend and AGPLv3 for the user interface, and its implications for proprietary agent development and community contributions.
📄 Licensing Clarifications: The LICENSES.md and README.md files have been updated with explicit explanations of the mixed-license model, emphasizing the permissive backend and copyleft UI.
🔑 UI Extensibility Licensing Note: A new licensing note has been added to the UI extensibility documentation, explicitly detailing the AGPL-3.0 implications for modifying and offering the UI as a network service.

Changed

🔄 Refined Ecosystem Model Documentation: The documentation for the ecosystem model has been significantly revised to provide a clearer and more in-depth explanation of the platform's collaborative approach and licensing strategy.
📝 Clarified Agent Behaviors: Extensive updates have been made across several agent documentation pages (Company Knowledge Agent, Instructed Assistant, Teachable Assistant, Retrieval Agent, Document Navigation Assistant, and MCP Tool Agent) to provide clearer descriptions of their purpose, workflows, capabilities, limitations, and setup instructions.
📝 Improved UI Extensibility Guide: The UI extensibility documentation has been substantially rewritten to offer more detailed insights into architectural foundations, custom service implementation, development workflows, deployment, and strategic value.
📝 Updated MCP Tools Usage in SDK: The SDK documentation for using MCP tools has been rephrased for better clarity, particularly regarding connection configuration and authentication modes.
📄 Minor Documentation Updates: Small adjustments and rephrasing have been applied to the Web Search and Environment Variables documentation pages for improved readability and consistency.
📝 API Tokens Documentation Enhancement: The English version of the API Tokens documentation has been updated to include the revoke endpoint in its lifecycle table.

[v0.290.10] - 2026-06-01 - Enhancing Project Governance: CLA and Contributor Guidelines

Added

✨ Contributor License Agreement (CLA): Introduced a formal Individual Contributor License Agreement (CONTRIBUTOR_AGREEMENT.md) to clarify intellectual property rights for contributions.
🚀 Automated CLA Checks: Implemented a new GitHub Actions workflow (cla.yml) to automatically verify CLA signatures for pull requests, streamlining the contribution process.
📄 Pull Request Template: Added a standard pull request template (pull_request_template.md) to guide contributors in providing necessary information, improving PR quality and review efficiency.

Changed

✍️ Updated Contributing Guidelines: Expanded CONTRIBUTING.md with a new section detailing the Contributor License Agreement process and instructions for signing it.
⚖️ Refined Licensing Notice: Updated the project's NOTICE file to provide a more general description of the mixed-license model, referencing LICENSES.md for full details.

[v0.290.9] - 2026-06-01 - Next-Gen Agents and Operational Excellence

Added

🦾 New Agent Blueprints: Introduced a suite of powerful, pre-built agent blueprints for diverse use cases:
- 📄 Document Intelligence Assistant: A highly configurable Retrieval-Augmented Generation (RAG) agent that answers questions from internal documents with citations, replacing the previous generic RAG agent with advanced capabilities including context-sufficiency guards and user/organization memory.
- 🤝 Company Knowledge Agent: An extension of the Document Intelligence Assistant that incorporates human-in-the-loop for user consent and bot-in-the-loop for expert consultation, capturing human replies as organizational knowledge.
- 🧑‍💻 Expert Coordinator Agent: Designed to relay questions to human experts via Slack or Teams, verify their responses, and store them as reusable organizational knowledge. Channel configuration is now managed directly within the agent's profile.
- 🧠 Instructed Assistant: A foundational chat assistant that follows plain-text system prompts for focused tasks like drafting, translation, or style-guided responses without document retrieval.
- 📚 Teachable Assistant: An agent that learns behavior from a few examples (few-shot prompting), making it ideal for tasks requiring specific response styles or formats.
- 🔍 Retrieval Agent: A headless, composable building block that performs document retrieval from a knowledge base without generating an answer, providing raw context for custom workflows.
- 🧭 Document Navigation Assistant: A routing agent that intelligently identifies the most relevant knowledge bases/namespaces for a user's question, confirms the selection with the user, and then delegates to a RAG agent.
- 🛠️ MCP Tool Agent: Enables agents to connect to external Model Context Protocol (MCP) servers and execute actions in other systems, such as creating tickets, querying databases, or sending messages.
⚙️ Automated PostgreSQL Maintenance: Integrated continuous weekly cleanup and monthly repack jobs for the Dagster PostgreSQL database, ensuring long-term operational stability by managing log growth and freeing disk space.
📄 Comprehensive Environment Variable Reference: Added a new, automatically generated documentation page listing all environment variables, their purpose, and consumer services, clarifying setup and configuration.
🌐 Self-Hosted Web Search: Documented the new privacy-focused web search feature, powered by a self-hosted SearXNG meta-search instance for Open-WebUI, including its default search engine list and customization options.
🔗 MCP Tooling for SDK Agents: Provided detailed SDK guidance on connecting custom agents to external MCP servers and invoking their tools, including McpClientConfig and various authentication modes.
💬 External Tool Connectivity Overview: Expanded the Agents overview documentation to include a new section explaining how agents can interact with external systems using the Model Context Protocol (MCP).

Changed

🔄 Updated Documentation Structure: Significantly reorganized the agent documentation, reordering blueprint pages for improved logical flow and discoverability.
📝 Refined German Translations: Numerous minor textual and stylistic improvements across German documentation for enhanced clarity and consistency.
🔑 Clarified Tenant Deletion Rules: Updated multi-tenancy documentation to state that any tenant can be deleted as long as at least one remains, clarifying the lack of special deletion protection for the initial startup tenant.
⚙️ Standardized Startup Tenant Variables: Renamed environment variables related to initial tenant configuration (e.g., AIHUB_DEFAULT_TENANT_NAME to AIHUB_STARTUP_TENANT_NAME) for clearer semantics.
🌐 Updated Network Egress Requirements: Removed Jina AI from the list of external service dependencies in the network requirements documentation.
📄 Improved Main Page Tagline: Updated the tagline on the main documentation landing page for better clarity and impact.
🔄 Refined Docs Sync Script: Improved the sync-docs.sh script to precisely control which README.md files are synchronized into the docs/6_code_deep_dive section, preventing unintended content.

Removed

🗑️ Deprecated Agent Documentation: Removed the outdated "RAG Agent" and "Expert Asking Agent" documentation, as their functionalities have been superseded and enhanced by the new agent blueprints.
🗑️ Removed Internal Tooling Documentation: Cleaned up documentation pages previously generated from internal .claude, .github/actions, and infra/deployment directories, streamlining the documentation scope.

[v0.290.8] - 2026-06-01 - Introducing a Mixed-License Model and Enhanced Contribution Guidelines

Added

📄 New NOTICE File: Introduced a new NOTICE file to formally outline the project's mixed-license model, providing clear attribution and a detailed license breakdown for different components.
👨‍💻 Author Information for Web Package: Explicitly added the author (bbv Software Services AG) to the packages/web/package.json metadata.

Changed

⚖️ Revised Project Licensing Model: The project now operates under a mixed-license model; packages/web and packages/backup are explicitly licensed under AGPL-3.0-or-later, while most other packages remain Apache-2.0 and sysadmin-* packages are proprietary.
📝 Updated Contribution Guidelines: The CONTRIBUTING.md file has been significantly updated to reflect the new mixed-license model, detailing that contributions are subject to the target package's license and requiring contributors to sign a Contributor License Agreement (CLA).
📅 Extended Copyright Information: Updated the root LICENSE file and relevant package README.md files to extend the copyright period to 2026 and specify bbv Software Services AG as the copyright holder.
🔗 Enhanced License Documentation: CLAUDE.md, LICENSES.md, packages/backup/README.md, and packages/web/README.md have been updated to accurately reflect and reference the detailed per-package licensing matrix and the new NOTICE file, ensuring greater transparency.

[v0.290.7] - 2026-05-29 - Introducing API Token Management and Documentation

Added

📄 New Documentation for API Tokens: Added comprehensive documentation guiding users on how to generate, manage, and use personal API tokens (sk-) for authenticating REST API requests, ensuring secure and long-lived access for scripts and integrations.

Changed

📄 Refined SonarQube Configurations: Updated SonarQube issue suppression rules for backup services (Neo4j, Valkey, Milvus) to accurately reflect intentional design patterns and improve the clarity of code quality analysis.

Refactor

🧹 Optimized Regular Expressions: Streamlined regex patterns in ClickHouse and PostgreSQL backup services by adopting \w shorthand and the re.ASCII flag, enhancing readability and precision.
⚡️ Enhanced Milvus Backup Selection: Improved the efficiency of Milvus backup selection logic by replacing sorted()[-1] with max(), leading to a more direct and performant approach for identifying the latest backup.

[v0.290.5] - 2026-05-28 - Dependency Security Enhancement

Security

🔑 Bolstered Dependency Security: Introduced a minimum version constraint for the starlette dependency (>=1.0.1) to mitigate a known vulnerability in earlier versions and enhance overall project security.

[v0.290.4] - 2026-05-27 - Development Environment Update

Changed

⚡️ Upgraded pnpm Package Manager: The monorepo's package management tool has been updated to pnpm version 10.20.0, enhancing dependency resolution and optimizing the overall development workflow.

Added

🦾 Backup Service SonarCloud Analysis: Enabled SonarCloud scanning for the backup package to improve code quality and maintainability.
📄 Backup Service SonarCloud Configuration: Introduced sonar-project.properties for the packages/backup service, configuring its code analysis.

Changed

⚙️ SonarCloud Rule Exclusions: Updated SonarCloud configuration to exclude a rule (Web:ItemTagNotWithinContainerTagCheck) for Notification/Item.vue to accommodate new semantic HTML structures.

Refactor

✨ Enhanced Accessibility for Input Components: Improved form accessibility across various components by adding explicit input-id attributes to input fields and for attributes to their corresponding labels. This includes:
- Agent Selector
- Chips Input
- Vector Store Input
- Knowledge Namespace Create and Edit Modals
- Memory Edit components
- User Settings Language Selector
🚀 Optimized Data Retrieval Performance: Improved the efficiency of identifying active items and filtering events by transitioning from array filter and includes methods to more performant find and Set.has operations, notably in:
- Tenant details page navigation
- Event list filtering
- Agent, Knowledge document, and Thread navigation composables
🧹 Codebase Modernization and Cleanup:
- Modernized asynchronous operations and error handling in core functionalities, including the application's health check and authentication callbacks, by adopting async/await syntax.
- Replaced window.location.origin and window.localStorage with globalThis for improved cross-environment compatibility in the OIDC client plugin.
- Standardized Node.js built-in module imports (node:url) in Nuxt configuration.
- Streamlined conditional checks with optional chaining syntax for better readability.
- Removed redundant or empty CSS selectors and unnecessary empty <style scoped> blocks from various web components to reduce bundle size and improve code clarity.
semantic Improved Semantic HTML Structure: Refined the HTML structure in several components for better accessibility and semantic correctness, including:
Converting non-interactive <label> elements to <p> tags for displaying static information (e.g., tenant ID, knowledge document upload location, memory details).
Transforming Notification Item from a div to a semantically correct <li> element, and its interactive area from a div with role="link" to a button, along with converting the Notification Overlay and main notifications list containers to <ul> elements.

[v0.290.2] - 2026-05-27 - Introducing Comprehensive User Role Management

Added

✨ New User Role Management API Endpoints: Introduced assign_role and revoke_role API endpoints within the UserController, enabling robust management of tenant-scoped user roles.
🦾 Service Layer for Role Actions: Implemented core service and persistence logic for assigning and revoking user roles, including validation for tenant membership and role existence.
🧪 Dedicated API Test Suite: Added a comprehensive test suite to ensure the reliability and correctness of the new user role assignment and revocation APIs.
🖼️ Interactive User Role Chips Component: Introduced a new UserRoleChips Vue component for the frontend, providing an intuitive interface to display, assign, and revoke user roles directly within user lists.
🚀 Frontend Composables for Role Management: Developed new composables (useAssignRoleToUser, useRevokeRoleFromUser) to streamline role management actions within the web application.
🌐 Internationalization Support for Role Management: Added new localization strings across supported languages to provide a consistent user experience for the new role management features.
⚙️ Centralized OpenAPI Schema Service: Introduced OpenApiSchemaService in packages/core to centralize and standardize the injection of tenant_id path parameters into OpenAPI schemas, improving SDK generation.

Changed

🔄 Integrated Role Management in User Controllers: The UserController in both the main API and the Sysadmin API now includes the new assign_role and revoke_role endpoints.
🤝 Shared OpenAPI Schema Generation: Both ApiRunner and SysadminApiRunner now leverage the new OpenApiSchemaService to ensure correct and consistent OpenAPI schema generation, especially for tenant-scoped paths.
📊 Enhanced User List Component: Updated the UserList component to seamlessly integrate the new UserRoleChips for interactive role display and management.
📦 Updated API Client SDK: The sysadmin-web and web API client SDKs have been regenerated to include the new role management methods and to accurately reflect tenant_id as a required path parameter for all tenant-scoped endpoints.
⚡️ Broader Cache Invalidation for Roles: Modified role-related mutations (create, update, delete) to trigger a broader cache invalidation, ensuring up-to-date data after changes.
📄 Documentation Updates: Revised CLAUDE.md and README.md in packages/sysadmin-api to reflect the adoption of OpenApiSchemaService and the extended functionality of the re-mounted UserController.

Refactor

🧹 Centralized OpenAPI Schema Logic: The logic for injecting tenant_id into OpenAPI schemas has been refactored from ApiRunner into the new OpenApiSchemaService for better modularity and reusability across runners.

Security

🔑 License Change for Web Package: The license for packages/web has been updated from Apache 2.0 to AGPL-3.0-or-later.

[v0.290.1] - 2026-05-27 - Enhanced Dependency Stability and Configuration Flexibility

Added

✨ Introduced strict peer dependency enforcement: A new .npmrc file has been added to enforce strict peer dependency rules, improving consistency and preventing potential dependency conflicts during development.
🚀 Refined Pnpm dependency overrides: New pnpm.overrides have been added for @vueuse/router to specifically pin vue-router to a compatible version, enhancing overall project stability.
📄 Configured Pnpm peer dependency allowances: Explicitly allowed specific versions of peer dependencies for key packages like @nuxt/schema, apexcharts, and magicast, addressing potential installation warnings or conflicts.

Removed

🗑️ Removed explicit package manager version lock: The packageManager field has been removed from the web package, allowing for more flexible package manager usage and reducing configuration overhead.

[v0.290.0] - 2026-05-27 - Multi-Tenancy Expansion and Frontend Monorepo Standardization

Added

🛡️ Introduced proprietary Sysadmin Plane: New packages/sysadmin-api (backend) and packages/sysadmin-web (frontend) services have been added for multi-tenant administration, now shipped under a commercial license.
📄 New LICENSES.md for explicit mixed-license model: A dedicated document now clearly outlines the licensing for each package (Apache-2.0, AGPL-3.0-or-later, and Proprietary terms), significantly enhancing legal clarity.
🐳 Dedicated Docker ignore file (.dockerignore): A new .dockerignore file has been added at the repository root to improve Docker image build efficiency and reduce image size by excluding unnecessary development and build artifacts.
🚀 New GitHub Actions workflow for Sysadmin Builds: A dedicated workflow (build-sysadmin.yml) automates the build and release process for the new sysadmin-api and sysadmin-web Docker images, ensuring consistent deployment.
🔑 Enhanced API get_my_identity endpoint: A lightweight user identity endpoint has been introduced in the API (/{tenant_id}/my-account/identity) that omits the full access matrix, making it suitable for quick sysadmin role checks across different API runners.
⚙️ Centralized Frontend Runtime Configuration: A new Nuxt plugin (0.runtime-config.client.ts) implements synchronous loading of runtime configurations (OIDC, WebSocket, sysadmin URL) from a global window.__AIHUB_CONFIG__ object, improving reliability and performance.
🤝 Centralized Frontend API Client Configuration: A new Nuxt plugin (api-client.client.ts) configures the SDK client globally, ensuring consistent token injection and error handling across the main and extended frontend applications.
📝 Architectural Decision Records (ADRs) for Sysadmin Plane: New ADRs document the rationale and consequences of splitting the sysadmin plane into separately licensed proprietary packages, ensuring transparency and future maintainability of architectural decisions.
💻 IDE Run Configurations for Sysadmin Services: Added new IntelliJ/VS Code run configurations to streamline local development, testing, and deployment of the sysadmin-api and sysadmin-web services.
🏷️ Automated License Annotation in Docker Compose: Generated Docker Compose files now include inline license comments for each service, improving license transparency directly within the deployment configuration.

Changed

📦 Frontend Monorepo Restructuring: The main frontend application's root directory has been refactored from packages/web/swiss_ai_hub_web to packages/web, simplifying paths and improving monorepo consistency.
🛠️ Updated Frontend Tooling Configuration: Paths and configurations across build scripts, ESLint, Dependabot, and SDK generation have been adjusted to align with the new packages/web structure and pnpm workspace setup.
🔒 Keycloak Integration for Sysadmin Plane: Keycloak realm configurations have been updated to support Single Sign-On (SSO) for the new sysadmin.${DOMAIN} subdomain, enabling seamless authentication between the main and sysadmin UIs.
🧹 Centralized .gitignore for Node.js modules: The exclusion of node_modules has been moved to the repository root .gitignore for better consistency and maintainability across the monorepo.
📖 Updated Documentation for Mixed Licensing: All relevant documentation (README, arc42, user guides) has been revised to reflect the new mixed-license model and refer to the LICENSES.md for detailed terms.
📦 Frontend package.json Updates: The package.json for the web package has been modified to reflect its new name (@swiss-ai-hub/web), new AGPL-3.0-or-later license, and alignment with the pnpm workspace.
⚙️ Frontend Docker Image Configuration: The Dockerfile and Nginx configuration for the web service have been updated, including health checks, changing the default port to 8080, and implementing the new runtime config loading mechanism (config.json to config.js).

Removed

🗑️ Deprecated TenantAdminController from Main API: The tenant administration API endpoints have been entirely migrated from the main packages/api to the new proprietary packages/sysadmin-api service.
🔄 Obsolete config-loader.client.ts plugin: This plugin, responsible for fetching runtime configuration, has been replaced by the new, more efficient 0.runtime-config.client.ts plugin.
🧹 Removed redundant Frontend pnpm-workspace.yaml: The pnpm workspace configuration for the frontend is now consolidated at the monorepo root.
🚫 Removed dedicated sysadmin middleware from packages/web: Its functionality has been moved to the new packages/sysadmin-web service to ensure proper confinement and role-gating within the proprietary plane.
🗑️ Deleted frontend config.template.json: This configuration template was replaced by config.template.js as part of the new runtime configuration loading mechanism.

Refactor

🧹 Centralized API Routes: Standardized and centralized the definition of API routes for agent, process, role, and tenant management endpoints, improving consistency and maintainability.
🔄 Enhanced Multimodal Content Handling: Refactored the internal logic for processing multimodal chat completion messages, leading to clearer, more robust content resolution for OpenAI API interactions.
⚡️ Modularized OpenAI Streaming: Split the OpenAI chat completion streaming event generation into dedicated, reusable helper methods for improved readability and easier maintenance.
📄 Standardized Error Messages: Centralized common error messages, such as "Not authorized to view this database," "Role not found," and "Tenant not found," ensuring consistent error responses across the API.
🧹 Constants for Code Clarity: Introduced dedicated constants for frequently used strings like UTC offset suffixes and i18n paths, enhancing code clarity and reducing magic string usage.

Changed

🩹 Improved Websocket Error Logging: Upgraded websocket disconnection error handling to include full traceback information, aiding in faster diagnosis and debugging.
🛡️ Centralized Role Service Error Handling: Shifted specific error handling for missing roles from controllers to the RoleService, streamlining API responses and improving error consistency.

Fixed

🐛 Corrected Model Creation Timestamps: Ensured that ModelDetails for OpenAI responses generate unique creation timestamps for each instance by using a default factory.

[v0.289.22] - 2026-05-26 - Secure MCP Tool Integration and Robust CI/CD Pinning

Added

🔑 Introduced User-Token Passthrough for MCP Clients: Agents can now securely connect to external Model Context Protocol (MCP) servers using the requesting user's bearer token, ensuring proper attribution and per-user authorization in external systems.
📄 New Documentation on MCP Client Authentication: Detailed guides and architecture decision records explain the new auth_mode options (none, api_key, user_token) for MCP connections.
🔒 GitHub Actions Pinning Enforcement: A new CI/CD job and configuration (pinact.yml) have been added to automatically verify and enforce that all GitHub Actions are pinned to specific commit SHAs, enhancing workflow security and stability.
🧪 Comprehensive Unit Tests for MCP Authentication: New test suites ensure the correct behavior and robust error handling of MCP client authentication mechanisms.

Changed

⚙️ Refined MCP Client Configuration: The McpClientConfig now includes an explicit auth_mode field, offering clear choices for authentication (none, API key, user token), with improved UI presentation for the API key.
🔄 McpReactAgent Updated for User-Token Authentication: The core McpReactAgent now leverages the new McpAuthResolver to pass user tokens to MCP servers when configured for user_token authentication.
📚 GitHub Actions README Updated: The documentation for GitHub Actions best practices has been updated to recommend pinning to 40-character commit SHAs for increased stability.

Refactor

🛡️ Pinned GitHub Actions to Specific SHAs: All GitHub Actions used in workflows and custom actions have been updated from major version tags to immutable 40-character commit SHAs, significantly improving the security and reliability of the CI/CD pipeline.
🧹 Improved MCP Client Factory Logic: The McpClientFactory has been refactored to support the new authentication modes, including explicit error handling for misconfigured connections (e.g., missing API key or user token).
⚡️ Frontend Composable Optimization: Several frontend composables have been refactored to streamline data fetching, removing redundant imports and simplifying their implementation.
🛠️ Internal Process Dispatcher Refinement: Logic for binding configuration to events within the process dispatcher has been extracted into a dedicated static method for better code organization.

[v0.289.21] - 2026-05-26 - Enhanced Workflow Security and Robustness

Refactor

🔒 Improved GitHub Actions Script Security: Refactored all CI/CD workflows to use explicit environment variables for GitHub event contexts, significantly enhancing script execution safety and mitigating potential command injection vulnerabilities.
🧹 Standardized Workflow Variable Handling: Streamlined the access of GitHub event inputs and context within workflow scripts, resulting in a cleaner, more consistent, and robust implementation for automated processes.

Added

🧪 Comprehensive NATS Sensor Tests: Introduced new unit tests for the NATS Document Uploaded Sensor, specifically for the event consumption logic, to ensure reliable and correct processing of messages.

Refactor

🧹 Improved S3 Error Logging: Switched to using logger.exception for directory existence and listing failures in the S3 Data Lake Client, providing more detailed stack trace information for easier debugging.
🔄 Streamlined SharePoint Retry Logic: Extracted asynchronous retry and backoff logic into a dedicated helper method within the SharePoint resource, enhancing code readability and maintainability.
⚡️ Modularized NATS Sensor Event Processing: Refactored the NATS Document Uploaded Sensor to abstract event consumption, validation, and acknowledgment into a new, dedicated function, leading to clearer, more robust event handling.

[v0.289.19] - 2026-05-26 - Improved Tenant-Aware Document Source Retrieval

Changed

🚀 Document Source Retrieval: Updated the generation of document source URLs to explicitly include the tenantId, ensuring accurate and tenant-specific access to original documents within a multi-tenant environment.

[v0.289.18] - 2026-05-22 - Enhanced Tenant Management and Robust Initialization

Added

🚀 Superuser Active Tenant Initialization: The active tenant for the superuser is now automatically ensured during the database initialization process, making the initial setup more robust and consistent.

Changed

⚡️ Improved Error Logging: Critical database initialization steps, such as role, bucket, and namespace creation, now utilize logger.exception to automatically include detailed stack trace information when errors occur, significantly improving diagnostic capabilities.
🔄 Refined Default Tenant Selection Fallback: The fallback logic for automatically selecting an active tenant has been updated to reliably choose an available tenant ID, ensuring a consistent active tenant assignment even under specific edge cases.

Refactor

🧹 Centralized Active Tenant Management Logic: The core logic responsible for automatically selecting and ensuring a user's active tenant has been moved from KeycloakAuthHandler to the KeycloakAdminService, enhancing modularity and centralizing Keycloak-related administrative functions.

[v0.289.17] - 2026-05-22 - OpenWebUI Upgrade and Backend Reliability Enhancements

Changed

⬆️ Upgraded OpenWebUI: Updated the integrated OpenWebUI platform to version v0.9.5, incorporating its latest features and stability improvements across all deployment configurations.

Fixed

⚡️ Improved UI Context Synchronization: Enhanced the reliability of UI context updates by switching from two-way event_caller to one-way event_emitter for JavaScript execution. This change prevents potential deadlocks that could occur in distributed OpenWebUI environments with Redis-coordinated session pools, ensuring smoother and more stable user experience.

Removed

🗑️ Cleaned Up Development Certificates: Removed legacy dev-cert.pem and dev-key.pem files from the Traefik configuration, streamlining the development environment setup.

[v0.289.16] - 2026-05-22 - Improved Pipeline Scheduling and Path Safety

Added

✨ Introduced run_after_success_sensor: A new Dagster sensor enabling jobs to be automatically triggered upon the successful completion of another job, streamlining pipeline workflows and ensuring proper sequencing.
🚀 New Development Workflow Commands: Added down-dev to easily stop the Docker Compose development environment and playground to quickly launch the Dagster pipeline SDK demo.

Changed

🔄 Refined Pipeline Job Scheduling: Cleanup (remove) jobs are now automatically triggered via a new sensor after their corresponding observation (observe) jobs succeed. This replaces their previous independent daily schedules, ensuring more timely and dependent cleanup operations.
📄 Updated Documentation for Pipeline Triggers: The CLAUDE.md documentation has been updated to describe the new run-status chaining mechanism for job automation.
🛡️ Improved Path Traversal Prevention: The create_figures_folder_name utility now includes enhanced validation to explicitly prevent path traversal attempts using . or .. as filenames, increasing system robustness.

Removed

🗑️ Deprecated remove_job_hour and remove_job_minute parameters: These scheduling parameters have been removed from various pipeline default_definitions as cleanup jobs are now dynamically triggered by the success of observation jobs.
🧹 Eliminated Redundant IDE Configuration: Removed the .idea/modules.xml file, streamlining repository setup for development environments.

[v0.289.15] - 2026-05-22 - Enhanced User Identity Resolution and Teams Bot Fixes

Fixed

🐛 Teams Bot User Lookup: Resolved a critical issue (#1314) where the Teams bot failed to authenticate users whose display name in Teams did not directly match their Keycloak email, by correctly resolving the user's true email via the Teams connector.

Added

✨ Centralized User Identity Resolution: Introduced new resolve_user_email and resolve_user_identity methods in the base CompletionHandler. This provides a robust and consistent mechanism to determine a user's email from various chat platforms (e.g., Teams) and resolve their Keycloak-backed identity across all bot types.
🧪 Comprehensive User Identity Tests: Added extensive unit tests for user email and identity resolution, covering various scenarios and edge cases to ensure the reliability and correctness of user authentication flows.

Refactor

🧹 Unified User Identity Logic: Consolidated the user identity resolution logic into the shared CompletionHandler, moving it from the AgentCompletionHandler and OpenaiCompletionHandler. This improves code reusability, maintainability, and ensures consistent user authentication across all bot implementations.

[v0.289.14] - 2026-05-22 - Enhanced AI Workflow Security

Security

🔒 Strengthened Claude Code Review Workflow Security: The automated Claude AI code review workflow is now restricted to run exclusively for pull requests from within the same repository and initiated by trusted contributors (Owners, Members, or Collaborators).
🔑 Improved Claude Bot Interaction Control: Access to trigger the Claude AI bot via comments or issues (using @claude) has been limited to repository Owners, Members, and Collaborators, enhancing control and preventing unauthorized invocations.

[v0.289.13] - 2026-05-22 - Securely Propagate User Identity to Agents

Added

🔑 Introduced X-AIHub-* Header Propagation: Implemented a new mechanism to securely propagate X-AIHub-* identity headers from incoming API requests through the NATS event bus to agent RunContext. This allows agents to perform actions on behalf of the originating user, supporting use cases like delegated authentication.
🔐 _aihub_headers Attribute to Base Events: Added a private attribute _aihub_headers to the BaseEvent class, enabling temporary storage of untrusted X-AIHub-* headers received from NATS messages for subsequent processing and validation by agent steps.
⚡️ New NATSMessageHeaders Utilities: Introduced with_aihub_headers for filtering and merging X-AIHub-* headers onto outgoing NATS messages, and extract_aihub_headers for safely extracting and standardizing these headers from incoming messages.
🧪 Comprehensive Unit Tests for Header Handling: Added new unit tests to ensure the robust and secure handling of X-AIHub-* headers across publishers and the NATSMessageHeaders utility.

Changed

🔄 API Request Header Extraction: Modified the OpenAI API controller to extract X-AIHub-* headers from incoming HTTP requests, ensuring they are captured early in the request lifecycle.
🚀 Event Distributor Header Forwarding: Updated the event distributor to forward X-AIHub-* headers onto NATS control-path events (StartEvent, HITL/BITL responses) while intentionally excluding them from observability-only display events to prevent credential over-sharing.
📦 Agent RunContext Identity Storage: Enhanced the AgentDispatcher to persist extracted X-AIHub-* headers into the agent's RunContext, making them accessible to agent steps that require user identity for delegated operations.
📡 NATS Publisher and Subscriber Integration: Integrated X-AIHub-* header processing into both NATS Core and JetStream publishers and subscribers, enabling seamless propagation and extraction of these headers across the event bus.

Changed

⚙️ Enhanced conversation timestamp accuracy by switching to timezone-aware UTC datetime objects for last_activity records, ensuring more consistent timekeeping.

Refactor

🧹 Streamlined content type checks within the content extractor, consolidating multiple startswith conditions for improved code readability.
⚡️ Optimized Slack ID retrieval by making the _get_slack_ids method synchronous, simplifying its execution flow.

Refactor

🧹 Streamlined Agent Dispatching Logic: Extracted event normalization and handling into dedicated helper methods within the AgentDispatcher for improved modularity, readability, and maintainability of event processing.
⚙️ Enhanced Multiprocess Runner Stability: Centralized and improved the graceful shutdown mechanism for MultiprocessAgentRunner to ensure more robust and consistent stopping of agent processes.
🔄 Simplified RAG Agent Preconditions: Refined the context_ready_for_history_limit precondition in the RAG agent by removing an unnecessary parameter, leading to cleaner code.
📄 Improved Module Import Clarity: Introduced module constants for lazy imports in the MCP module, enhancing code readability and reducing string literal duplication.

Changed

📝 Clarified Event Tracing Behavior: Added a comment to explain that EventDisplayer arguments are intentionally excluded from tracing, improving code documentation.

[v0.289.10] - 2026-05-21 - Workflow Optimization and Release Tagging Enhancements

Fixed

🔑 Resolved Release Tagging Permissions: Addressed an issue in the set-latest workflow that prevented latest git tags from being moved when workflow files were modified, by leveraging an SSH deploy key for proper authentication.

Refactor

🧹 Streamlined CI/CD YAML Parsing: Replaced Python-based YAML parsing with yq and jq in the build-agents, build-pipelines, and set-latest workflows. This change utilizes pre-installed tools on the ubuntu-slim runner, enhancing workflow reliability and reducing external dependencies.

[v0.289.9] - 2026-05-20 - Infrastructure Maintenance

Changed

⚙️ Updated Playwright Docker image to v1.60.0: Upgraded the underlying browser automation engine to leverage the latest features, bug fixes, and performance improvements for services utilizing Playwright.

[v0.289.8] - 2026-05-20 - Optimized CI/CD Workflows for Enhanced Performance

Refactor

🚀 Upgraded CI/CD Runners: Migrated a wide range of GitHub Actions workflows from generic ubuntu-latest to more specialized and performant runners like ubuntu-slim and ubuntu-24.04-arm. This optimization targets improved execution speed, resource efficiency, and broader compatibility across various build and test environments.
⚙️ Enhanced Test Module Flexibility: Introduced support for matrix-based runner selection in the Test Modules workflow, enabling dynamic allocation of testing environments for different modules. This change also transitions testing for these modules from a dedicated ubuntu-latest-8-cores instance to a more standard ubuntu-latest (via matrix configuration), paving the way for future environment customizations.
⚡️ Accelerated Build and Release Processes: Key workflows such as release creation, changelog generation, license reporting, environment documentation, and documentation deployment are now running on ubuntu-24.04-arm, resulting in faster completion times.
🧹 Streamlined Development Feedback Loops: Linting, branch/PR semantic checks, and AI-assisted code review workflows have been moved to ubuntu-slim runners, providing quicker feedback on code quality and PR readiness.

[v0.289.7] - 2026-05-20 - Improved Workflow Resilience with Configured Timeouts

Changed

⚡️ Enhanced CI/CD Job Stability: Introduced explicit timeout-minutes configurations across numerous GitHub Actions workflows, ensuring that jobs terminate gracefully and preventing prolonged or stuck executions.

[v0.289.6] - 2026-05-20 - Enhanced LLM Ecosystem and Robust Dependency Management

Added

📦 Introduced pnpm-workspace.yaml configurations across frontend and documentation projects, leveraging pnpm's native minimumReleaseAge for enhanced supply-chain security and consistent dependency management.
⚙️ Added onlyBuiltDependencies configurations for selected frontend packages to ensure compatibility with pnpm 10's stricter default behavior regarding dependency build scripts.
🧪 Implemented a new integration test suite for LiteLLM text generation models, dynamically verifying the functionality and responsiveness of all advertised models on the Swiss LLM Cloud proxy.

Changed

🧠 Updated the default text generation LLM for agents and pipelines to text-generation/gemma-4-31B-it, replacing the retired gpt-oss-120b for improved performance and access to the latest models.
🚀 Expanded the Swiss LLM Cloud model catalog in LiteLLM, incorporating new models such as Kimi-K2.6, Ministral-3-14B-Instruct-2512, and Qwen3.5-122B-A10B-FP8, and updating cost metrics for gemma-4-31B-it.
🔄 Enhanced CI/CD frontend linting workflows to utilize corepack for pnpm installation, ensuring a more consistent and robust dependency setup across development and automated testing environments.
⬆️ Upgraded the pinned pnpm package manager version to 10.20.0, contributing to improved build process stability and security.

Fixed

🐛 Resolved Dependabot CI failures stemming from pnpm incompatibilities with Dependabot's cooldown mechanism by transitioning to pnpm's native minimumReleaseAge feature.
📄 Corrected and updated documentation across README.md and arc42 decisions to accurately reflect the latest available LLM models on the Swiss LLM Cloud and their usage in the platform.

[v0.289.5] - 2026-05-19 - Refined Dependabot Cooldowns

Changed

🔄 Streamlined Dependabot Cooldowns: Corrected Dependabot configurations for Docker and GitHub Actions ecosystems by removing unsupported semver-minor-days and semver-patch-days cooldown settings. These ecosystems exclusively support default-days, as their tags are not semver-decomposable, ensuring more accurate and effective dependency update management.

[v0.289.4] - 2026-05-19 - Fortified Repository with Codeowners and Tuned Dependabot

Security

🔑 Implemented CODEOWNERS for Critical Paths: Enforced mandatory reviews from core maintainers for changes to critical areas like CI/CD, deployment infrastructure, shared core packages, and security policies, significantly enhancing repository security and integrity.

Changed

⚙️ Refined Dependabot Update Strategy: Overhauled Dependabot configuration to improve stability and reduce PR noise by switching to monthly update intervals, introducing cooldown periods for new releases, and intelligently grouping version and security updates.
📄 Updated Contribution Guidelines: Added clear guidance to CONTRIBUTING.md outlining the new CODEOWNERS review requirements for sensitive parts of the codebase.

Refactor

🧹 Adjusted Dockerfile Path in Dependabot: Corrected the directory reference for the Playwright Docker deployment within the Dependabot configuration.

[v0.289.3] - 2026-05-12 - Dynamic Forms and Stability Enhancements

Fixed

🐛 Fixed Number Input Validation: Ensured FormKit validation for InputNumber elements correctly handles large and small decimal bounds by preventing scientific notation in validation rules.
🐛 Corrected ChipsInput Required Logic: Prevented ChipsInput elements from being automatically marked as required, acknowledging that an empty list is a valid "unset" state for list-collecting inputs.
🚫 Hard-Pinned Nuxt.js to 3.21.0: Temporarily reverted Nuxt.js to version 3.21.0 to mitigate a vite-node IPC regression that caused hangs and module resolution errors in development mode.

Added

✨ Introduced FormKitDynamicConfiguration Component: Added a new, reusable Vue component to centralize dynamic form rendering logic, including repeater handling and locale string support, for consistent UI across agent and process configurations.
🦾 New seedFormDefaults Helper: Implemented a new composable function to recursively apply backend Pydantic default values to form data, ensuring fields are correctly pre-filled upon initialization.
🧪 Expanded Form Validation Tests: Added comprehensive unit tests for InputNumber validation bounds (integers, floats, large, and small decimals) and verified ChipsInput requirement logic, improving form reliability.
📄 Documented Nuxt Hard-Pin Decision: Added an Architecture Decision Record (ADR) explaining the rationale, consequences, and exit criteria for hard-pinning Nuxt.js to version 3.21.0.

Refactor

🔄 Streamlined Agent and Process Configurations: Replaced internal form handling logic in Agent/Configuration.vue and Process/Configuration.vue with the new reusable FormKitDynamicConfiguration component, reducing code duplication.
⚡️ Enhanced Dynamic Form Stability: Implemented preserve: true and unique key attributes across FormKit groups and inputs to prevent data loss and improve rendering stability during conditional display or re-rendering.
🧹 Optimized Form Instance Data Initialization: Refined the data hydration and default seeding logic when creating new agent or process instances, ensuring consistent and correct form state.

[v0.289.2] - 2026-05-12 - Refined Backup Orchestration and CI Scopes

Fixed

🐛 Dagster Backup Asset Dependencies: Corrected the execution order of Dagster backup maintenance assets to ensure the postgres_autovacuum_tune process runs after postgres_indexes, preventing potential deadlocks and enhancing the reliability of backup workflows.

Changed

⚙️ CI/CD Semantic Commit Scope: Added backup as an allowed scope for semantic commit messages within CI/CD workflows, enabling clearer categorization and validation of changes related to the backup system.
📄 Developer Documentation: Updated the CLAUDE.md documentation to reflect the newly introduced backup scope for semantic commit messages, guiding developers on proper commit conventions.
🧪 Backup Maintenance Tests: Updated unit tests to accurately validate the corrected dependency ordering of backup maintenance handlers, including the new dependency of postgres_autovacuum_tune on postgres_indexes.

[v0.289.1] - 2026-05-12 - Dependency Management & Stability Enhancements

Changed

📄 Streamlined Dependabot configuration by excluding Python base images from automatic major version bumps, allowing for more deliberate and controlled updates of core environments.
🔄 Relaxed the Llama-index-core dependency constraint in the core package to permit compatible minor version updates, providing greater flexibility for patch releases.
⬆️ Updated the Hugging Face Hub dependency for development tools, streamlining its integration and removing an unnecessary [cli] extra.

Fixed

🐛 Introduced a pnpm override for @ungap/structured-clone across documentation and web packages to ensure specific, stable versions of transitive dependencies are utilized, enhancing overall stability.

[v0.289.0] - 2026-05-12 - Next-Gen Memory Scoping and Agent Templates for Expert Agents

Added

✨ Agent Templates: Introduced a new templating system for ExpertAskingAgent and ExpertRAGAgent, making it easier to deploy and configure specialized agents for various use cases.
🦾 Engineering Expert Agent Templates: Shipped default Engineering Expert and Engineering Expert RAG agent templates, providing ready-to-use configurations for escalating and answering technical questions.
⚙️ Granular Organization Memory Scoping: Added new configuration objects (OrgMemoryReadConfig, OrgMemoryWriteConfig) allowing agents to define default_tenant_namespace and allowed_tenant_namespaces for precise control over knowledge access.
📝 Configurable Organization Memory Format: ExpertAskingAgent now supports a configurable org_memory_format template, enabling custom Q&A snippets to be stored in organization memory.
🎛️ New Form Components: Introduced ChipsInput for intuitive list input (e.g., allowed namespaces) and OrgMemoryTenantInput which integrates built-in authorization checks for organization memory fields directly into the UI.
🔑 Authorization for Organization Memory: Implemented client-side configuration authorization checks for organization memory settings, ensuring users only configure resources they have access to.
📄 Detailed Environment Variable Descriptions: Added comprehensive descriptions and default values for many backup, Milvus, NATS, Redis, and S3 storage-related environment variables in the deployment guide, improving clarity.

Changed

🔄 Refined Organization Memory APIs: Updated internal organization memory APIs and event structures to support granular namespace lists, moving from a single tenant_namespace string to a list of tenant_namespaces for search and a specified singular namespace for writes.
🧹 Streamlined Reranking Configuration: Simplified the configuration for document reranking in RAG agents, making it more straightforward to enable or disable the feature.
📋 Environment Variable Renaming for Consistency: Renamed several backup-related environment variables (e.g., DAGSTER_DB to BACKUP_DAGSTER_DB) to ensure consistent naming conventions across the platform and documentation.
🖼️ Vector Store UI Improvement: Refactored the "Allowed metadata filter fields" input in the vector store configuration to utilize the new ChipsInput component for a more user-friendly experience.

Refactor

🗑️ Consolidated Memory Configuration: The MemoryConfig class has been removed and its functionality split into UserMemoryConfig and OrgMemoryReadConfig/OrgMemoryWriteConfig for clearer separation and management of user-scoped and organization-scoped memory settings.
⚙️ Internal .gitignore Enhancements: Added .claude/*.lock to .gitignore to prevent unnecessary lock files from being tracked.

[v0.288.1] - 2026-05-11 - Streamlined Environment Configuration and Validation

Added

📄 Introduced a comprehensive, auto-generated environment variables reference page in the documentation, providing a single source of truth for all configurable settings.
🚀 Implemented automated consistency checks for environment variables against Docker Compose configurations and Pydantic settings during PR analysis and release workflows, enhancing deployment reliability.
🔑 Integrated Keycloak OAuth2 Proxy client for the Backup Dagster UI, enabling secure, centrally managed access to backup management interfaces.
💾 Added explicit support for Azure Data Lake Storage, Azure Document Intelligence, and SharePoint by introducing dedicated environment variables for their configurations, facilitating data ingestion from these sources.
🔐 Introduced new internal environment variables for NATS and Redis/Valkey tokens, improving security and granular configuration control for these messaging and caching services.
👷‍♀️ New check-env and generate-env-docs Makefile targets for manual execution of environment validation and documentation generation.

Changed

📝 Reorganized the .env.dev file structure, moving host-side development overrides to a dedicated section for improved clarity and maintainability.
🔄 Standardized OAuth2 Proxy provider display names to "Keycloak" across various services in Docker Compose templates, ensuring a consistent user experience.
⚙️ Made Mem0 model names fully configurable via environment variables (MEM0_LLM_NAME, MEM0_EMBEDDING_MODEL_NAME, MEM0_RERANKING_MODEL_NAME) in Docker Compose, offering greater flexibility for AI model selection.
🔄 Updated the default AIHUB_STARTUP_TENANT_ID from 'swiss_ai_hub' to 'default' for a more generic initial platform setup.
🔒 Hardcoded the Langfuse NEXTAUTH_URL directly within Docker Compose templates based on the deployment stage (dev/prod), simplifying configuration by removing reliance on an .env variable for this specific setting.
📈 Fixed pgbouncer pool size values to constant values within Docker Compose, removing optional environment variable overrides for these specific parameters.
📚 Improved documentation comments within .env.dev to more clearly explain sections and variable usage.
🧹 Added root-level Ruff configuration to pyproject.toml for consistent code formatting across repository-level scripts.

Removed

🗑️ Removed the redundant WEBUI_SECRET_KEY environment variable, as its functionality is now consolidated under OPENWEBUI_SECRET_KEY.
❌ Eliminated deprecated fake authentication variables (DANGEROUS_DEV_ONLY_AUTH_FAKE_*) previously used for development-only testing.
🗑️ Removed the explicit AIHUB_CREATE_DEFAULT_ROLES environment variable, as role creation now defaults to True within the application code.
🗑️ Removed OAUTH_AUTHORITY_URL from .env.prod.

Refactor

🧹 Centralized environment variable management logic into new Python modules (env_check.py, env_docs.py, env_inventory.py) to streamline validation, documentation generation, and overall configuration handling.

[v0.288.0] - 2026-05-11 - Streamlined Optional Form Fields and Configuration Management

Refactor

🔄 Revised Form Optionality Handling: Overhauled how optional form fields are managed, replacing explicit enabled: bool flags with Pydantic's T | None annotations for improved data model clarity and consistency. This architectural change is documented in 2026_05_07_nullable_form_fields_over_enabled_toggle.md.
🦾 Unified UI Toggle Management: Integrated UI toggle logic directly into the form framework, automatically generating enable/disable checkboxes for nullable fields and groups. This standardizes the user experience and significantly reduces boilerplate for optional settings.

Changed

⚡️ Simplified Agent Reranking Configuration: Updated RAG and Expert RAG agents to leverage the new nullable field system, removing the dedicated enabled field from RerankingConfig and simplifying internal logic to check for is not None.
🧹 Cleaned Agent Configuration Data: Agent configurations now store null for disabled optional sub-forms, leading to smaller, more readable persisted data by eliminating stale default values.
🚀 Enhanced Frontend Form Data Processing: Introduced new seedNullableToggles and coerceNullableToggles functions to intelligently initialize and submit form data, correctly managing the state of auto-generated nullable field toggles in the UI.

Removed

🗑️ Eliminated Redundant enabled Fields: The dedicated enabled boolean fields in RerankingConfig and other optional sub-configurations have been removed, as their functionality is now implicitly handled by the form framework based on T | None annotations.
📄 Deprecated Reranking UI Conditions and Translations: Removed condition_if properties that previously linked to the reranking_config_enabled field, along with their associated internationalization strings, simplifying the UI schema.

[v0.287.3] - 2026-05-11 - Improved Memory Management API

Refactor

🧹 Streamlined Memory Operations: Refactored memory update and delete functions to automatically derive the tenantId from the application context, simplifying API calls and improving consistency for developers.

[v0.287.2] - 2026-05-11 - RAG Agents: Richer Context with Image Support

Changed

🖼️ Enhanced Context Sufficiency Guard: RAG agents now pass the full ChatMessage objects, including images and other rich content, to the context sufficiency guard, allowing Large Language Models (LLMs) to process multimodal information more effectively.
💬 Improved Chat History Forwarding: The chat history sent to the context sufficiency guard now retains its full ChatMessage structure, enabling more accurate and rich conversational context for LLMs.

Refactor

🧹 Updated Guard Prompting Mechanism: Switched the internal context sufficiency guard from PromptTemplate to RichPromptTemplate to support dynamic rendering of multimodal ChatMessage objects within LLM prompts.
⚙️ Standardized Context Parameter: Refactored the do_context_sufficient_guard function across RAG agents to consistently accept a ChatMessage object for context, rather than a flattened string.

[v0.287.1] - 2026-05-11 - Optimized Vector Store Queries with Namespace Scoping

Changed

⚡️ Optimized Vector Store Node Retrieval: Enhanced the efficiency of node retrieval across the platform by propagating namespace information down to the vector store. This change, applied in the Knowledge Service, Parent Summary Post-Processor, and Vector Previous/Next Post-Processor, ensures that queries are scoped to specific data partitions, significantly improving performance and data isolation in multi-tenant environments.
🔄 Updated PartitionAwareMilvusVectorStore: The get_nodes method in the PartitionAwareMilvusVectorStore now explicitly accepts a namespaces argument, allowing for more precise and efficient querying within designated data partitions.

Added

✅ Introduced Namespace Filter Propagation Tests: Added new unit tests to verify the correct forwarding and application of namespace filters within vector store post-processors, ensuring the reliability of partition-aware data retrieval.

[v0.287.0] - 2026-05-11 - Enhanced Memory Configuration for RAG Agents

Added

✨ Granular Memory Reranking Control: Introduced new configuration options to enable or disable reranking specifically for both organization memory and user memory searches. This provides greater control over retrieval relevance and performance, allowing users to optimize for latency or accuracy as needed.

Refactor

🧹 Consolidated RAG Agent Memory Configuration: Centralized all memory-related settings (including organization memory enablement, user memory retrieval/storage, tenant ID, and namespace) into a new, dedicated MemoryConfig object. This significantly streamlines agent configuration, improves readability, and simplifies future enhancements.
🔄 Updated Agent Memory Access: RAG agents and their preconditions now access all memory-related settings through the consolidated MemoryConfig object, ensuring consistent and organized handling of memory configurations across the system.

[v0.286.8] - 2026-05-08 - Internal Maintenance

[v0.286.7] - 2026-05-08 - Streamlined OAuth Configuration and Enhanced Tenant Setup

Added

⚙️ Introduced Startup Tenant Configuration: New environment variables (AIHUB_STARTUP_TENANT_ID, AIHUB_STARTUP_TENANT_NAME, AIHUB_STARTUP_TENANT_DESCRIPTION, AIHUB_STARTUP_TENANT_ACCESS_RULES) have been added to support initial setup and management of a default "startup tenant", providing more granular control over Keycloak realm integration.

Changed

🔄 Standardized OAuth Authority URL Configuration: The OAUTH_AUTHORITY_URL is now dynamically configured based on the deployment stage. It automatically points to http://localhost:8180/realms/aihub for development environments and https://auth.${DOMAIN}/realms/aihub for all other stages, simplifying setup and reducing manual configuration.

[v0.286.6] - 2026-05-08 - Enhanced Backup Robustness and Configurability

Added

✨ Configurable PostgreSQL Backup Timeout: Introduced BACKUP_POSTGRES_SUBPROCESS_TIMEOUT_SECONDS (defaulting to 6 hours) to allow operators to adjust the timeout for PostgreSQL dump and restore operations, preventing failures on large databases.
⚡️ NATS Subprocess Retry Mechanism: Implemented retry logic for NATS CLI commands within the backup process, enhancing resilience against transient connection issues or server restarts.

Changed

🔄 Standardized Backup Service Environment Variables: All environment variables specific to the backup service are now explicitly prefixed with BACKUP_ in Docker Compose configurations for clearer separation and identification.
🦾 Improved NATS Readiness Probing: The NATS readiness check now probes JetStream's stream list command instead of a basic RTT, providing a more accurate assessment of NATS service availability before proceeding with backups.
📄 Updated Backup Container Exclusion List: Added oauth2proxy services to the list of containers that remain running during a backup, ensuring the backup Dagster UI remains accessible through OAuth.

Fixed

🐛 Resolved PostgreSQL Backup Timeout Issues: Addressed scenarios where PostgreSQL backups of large databases could time out due to a previously fixed 5-minute subprocess limit, which is now configurable via the new BACKUP_POSTGRES_SUBPROCESS_TIMEOUT_SECONDS environment variable.
🐛 Mitigated Transient NATS Connection Failures: Enhanced NATS operations with retry logic, reducing backup failures caused by temporary NATS server unavailability during runs.

Refactor

🧹 Streamlined Internal Backup Settings Handling: The internal BackupSettings class now automatically handles the BACKUP_ prefix for environment variables, improving consistency and reducing boilerplate in the configuration.

[v0.286.5] - 2026-05-06 - Enhanced Dependency Update Management

Changed

🚫 Disabled Automatic Major Dependency Updates: Configured Dependabot to explicitly ignore major version bumps across all dependencies, ensuring that significant upgrades are handled with deliberate, planned migrations rather than automatically.
🔄 Streamlined Dependabot Grouping and Logic: Consolidated Python and Web dependency update groups and standardized grouping patterns across all ecosystems for minor and patch updates, reducing PR noise and clarifying update bundles.
📝 Enhanced Dependabot Configuration Documentation: Updated comments within the dependabot.yml file to provide clearer explanations of Dependabot's grouping behavior, uniqueness enforcement, and the updated major version update policy.

[v0.286.4] - 2026-05-06 - Optimized Milvus Vector Store for Scalable RAG

Added

🧪 Unit Tests for Filter Logic: Introduced comprehensive unit tests for the namespace filter extraction logic within the PartitionAwareMilvusVectorStore, ensuring its correctness and robustness for partition-aware operations.

Changed

🚀 Enhanced Milvus Partition Loading: Optimized the PartitionAwareMilvusVectorStore to intelligently load only necessary Milvus partitions during add, query, get_nodes, and delete operations. This significantly reduces memory consumption and improves performance for large-scale RAG systems by avoiding full collection loads.

Fixed

🐛 Robust Namespace Filter Extraction: Addressed an issue in the PartitionAwareMilvusVectorStore where nested MetadataFilters might not have been correctly parsed, potentially leading to inefficient full collection loads during queries. Namespace extraction is now more robust, ensuring precise partition targeting.

Refactor

🧹 Project Structure Modularization: The IntelliJ IDEA project configuration has been updated to reflect a more modular structure, with separate module files for different components like agent, api, core, and pipeline.

[v0.286.3] - 2026-05-05 - Build Process Modernization and Dependency Streamlining

Changed

⚡️ Updated Node.js Base Image: Migrated the Docker build environment from the Long-Term Support (LTS) Node.js version to the latest Node.js v25, enabling access to current features and performance enhancements.

Refactor

🧹 Modernized Corepack and pnpm Setup: Re-engineered the Dockerfile's dependency management strategy to explicitly install Corepack and leverage its corepack install command, ensuring the use of the exact pnpm version defined in package.json for improved consistency.
🔑 Improved Build Environment Security: Enhanced the build process by pinning Corepack to a specific version and removing potential conflicts from pre-shipped Yarn shims, contributing to a more secure and reliable build.

[v0.286.2] - 2026-05-05 - Streamlined Release Promotion and Smart Versioning

Added

✨ New Release Promotion Workflow: Introduced a dedicated GitHub Actions job to explicitly promote pre-releases to full "latest" status, ensuring consistency across Git tags, Docker images, and GitHub Releases.

Changed

🚀 Phased GitHub Releases: New GitHub releases are now initially marked as pre-releases, allowing for a controlled testing period before official promotion to "latest."
🔍 Smarter Installer Version Resolution: The install.sh script now intelligently distinguishes between official "latest" releases and pre-releases, prioritizing promoted versions while gracefully falling back to the most recent pre-release if no official "latest" is yet available.

[v0.286.1] - 2026-05-05 - Enhanced Build Stability and Dependency Optimization

Changed

🛡️ Strengthened Agent Build Validation: The CI/CD pipeline now enforces stricter validation for agent builds. All individual agent builds within a matrix must successfully complete for the entire workflow to pass, preventing silent partial failures.

Fixed

🐛 Resolved apache-age-python Dependency Issue: Eliminated an unnecessary transitive dependency on apache-age-python (pulled by mem0ai[graph]), which previously caused build complications due to its psycopg2 requirement for system libraries. This improves build reliability and streamlines dependency management.

[v0.286.0] - 2026-05-04 - Infrastructure Modernization and Dependency Upgrades

Changed

🛠️ Modernized CI/CD Infrastructure: Upgraded core GitHub Actions, including checkout, setup-python, setup-node, and artifact management, to enhance build reliability, performance, and security across all workflows.
🚀 Advanced Platform Technologies: Updated Node.js runtime versions across development environments and Docker images, providing a more current and efficient foundation for frontend services and CI processes.
✨ Enhanced Web Application Dependencies: Performed comprehensive updates to key frontend libraries and frameworks, including significant version bumps for FormKit and Pinia Colada, ensuring the web application benefits from the latest features and stability.
🧩 Improved Development and Testing Toolchain: Upgraded essential development dependencies such as pytest, ruff, and Playwright, leading to a more robust testing framework and optimized code quality checks.
📄 Modernized Documentation Dependencies: Updated various documentation-related packages like vitepress, mermaid, and cytoscape to ensure the documentation platform remains current and performant.

Refactor

🔄 Refined Test Mocking: Enhanced the flexibility of aiohttp request mocks in bot playground tests to improve test resilience and adaptability to evolving API interactions.
🔒 Standardized Dependency Constraints: Applied a precise version constraint for the mem0ai[graph] dependency to ensure compatibility and prevent unintended breaking changes from future major releases.

[v0.285.4] - 2026-05-04 - Enhanced Authentication Robustness and Performance

Fixed

🐛 Resolved OpenWebUI authentication issue (Issue #1050): Fixed a bug where authentication failed for proxied users if the underlying bearer token was owned by a service account without tenant membership, even when the proxied user had valid tenant associations. Token verification is now correctly decoupled from the token owner's tenant context.
✅ Added comprehensive regression tests: New end-to-end tests for the OpenWebUI authentication handler confirm the fix for issue #1050, and additional unit tests for TokenAuthHandler ensure robust token verification.

Changed

⚡️ Optimized user and role retrieval: Keycloak calls for fetching user details and realm roles in TokenAuthHandler are now executed concurrently using asyncio.gather, improving authentication performance.

Refactor

🧹 Decoupled token verification: The authentication process has been refactored to clearly separate the initial validation of a bearer token from the subsequent resolution of the associated user's details and tenant context, enhancing modularity and robustness.
🔄 Renamed BearerAuthHandler to TokenAuthHandler: The core bearer authentication handler has been renamed for improved clarity and consistency across the codebase.

[v0.285.3] - 2026-05-04 - Refined Image Management for Deployment Builds

Refactor

🧹 Streamlined Image Configuration: Simplified the compose-config.yml by removing redundant build: localbuild directives and stage-specific image tags for project-managed images like Postgres and Playwright. These images are now consistently referenced by their pinned, pre-built tags across all deployment stages.
🔄 Enhanced Image Tagging Automation: Refined the internal logic for identifying and tagging images during releases, ensuring that only actively rebuilt application images (those marked for per-release publishing) are updated with the :latest tag. Project-managed base images with specific version pins are now explicitly excluded from this automatic retagging process.
📦 Reorganized Playwright Dockerfile: Moved the Playwright Dockerfile to a more logical and standardized location within the infra/deployment/docker directory, improving project structure and maintainability.
⚙️ Simplified Docker Compose Generation: The internal templates for generating Docker Compose files have been updated, removing conditional build instructions for Postgres and Playwright services. These services now uniformly pull their pre-built, version-pinned images from the registry.

Changed

📄 Clarified Postgres Image Documentation: Updated the documentation for the Postgres image within CLAUDE.md to explicitly describe it as a project-managed image that is published out-of-band and consumed with a pinned tag, providing clearer guidance on its lifecycle.

[v0.285.2] - 2026-05-04 - Enhanced Release Workflow for Protected Branches

Changed

⚙️ Streamlined Release Finalization: The add-tag.yml GitHub Actions workflow now utilizes a dedicated deploy key for repository checkout. This enhancement ensures the workflow can successfully amend merge commits and force-push to the protected main branch, improving the reliability and integrity of the release finalization process.

[v0.285.1] - 2026-05-04 - Automated Dependency Management and CI/CD Streamlining

Added

✨ Automated Dependency Management: Introduced Dependabot configuration to automatically detect and create pull requests for dependency updates across Python, Node.js (frontend, docs, E2E), Docker base images, and GitHub Actions, significantly enhancing project maintainability and security.

Changed

📄 Improved Documentation: Updated the description for the github_token input in the backend linting action, clarifying its specific use by reviewdog for posting review comments.

Refactor

🧹 Optimized Pull Request Checks: Configured semantic PR checks to bypass title linting and version label requirements specifically for Dependabot-generated pull requests, streamlining the process for automated dependency updates.

Removed

🗑️ Streamlined CI/CD Workflows: Eliminated explicit Git SSH key configuration steps from various GitHub Actions workflows, simplifying pipeline setup and reducing configuration overhead.

[v0.285.0] - 2026-04-29 - Introducing Self-Hosted Web Search (SearXNG) and Playwright Update

Added

✨ Self-Hosted Web Search with SearXNG: Integrated a self-hosted SearXNG meta-search instance, providing privacy-aware web search capabilities directly within the platform. This offers greater control, eliminates per-query costs, and ensures auditable egress for web queries.
📄 Comprehensive Web Search Documentation: Added a new, detailed documentation guide explaining how Open-WebUI leverages SearXNG for web search, including the curated list of privacy-first search engines, customization options, and instructions on how to disable the feature.
🔑 New SEARXNG_SECRET_KEY Environment Variable: Introduced a new environment variable to secure the self-hosted SearXNG instance, enhancing operational security.

Changed

🔄 Switched Web Search Provider to SearXNG: The platform's web search functionality now exclusively uses the integrated SearXNG service, transitioning away from the previous Jina AI integration.
🌐 Updated Network Requirements for Web Search: The documentation for network requirements has been updated to reflect the new set of upstream search engines (e.g., Brave, DuckDuckGo, Mojeek, Qwant, Startpage, Wikidata, Wikipedia) queried by SearXNG.
🚀 Playwright Updated to v1.58.0: Upgraded the Playwright browser automation library to version v1.58.0, ensuring compatibility with the latest browser features and security updates for web browsing services.

Removed

🗑️ Removed Jina AI Integration: The dependency on Jina AI for web search has been entirely removed from the platform, including its associated API key configuration.

Changed

✨ Enhanced OpenAPI Tag Generation: The API documentation (Swagger UI) now comprehensively includes tags for all mounted controllers and intelligently prevents duplicate tag entries, providing a more complete and accurate overview of available API endpoints.
🌍 Localized API Endpoint Tags: Switched API endpoint tags from hardcoded strings to dynamically generated, internationalized strings, significantly improving multi-language support and consistency for API documentation.

[v0.284.1] - 2026-04-28 - Optimized LiteLLM Logging for Pipeline Efficiency

Added

✨ Introduced pipeline LLM message redaction: Significantly reduces LiteLLM spend log database size by redacting message content from pipeline-originated LLM calls (e.g., embeddings) while preserving essential cost and token tracking. This prevents database bloat from high-volume, reproducible pipeline traffic.
📄 Documented LiteLLM message redaction decision: Added an Architecture Decision Record (ADR) detailing the rationale, drivers, and consequences of implementing pipeline message redaction.

Changed

🚀 Upgraded LiteLLM proxy version: Updated the LiteLLM proxy to v1.83.10-stable to enable the new message redaction header and incorporate other upstream improvements.
🔄 Enhanced LLM configuration flexibility: Core LLM and Embedding Model configurations now support injecting custom HTTP headers, enabling fine-grained control over LiteLLM proxy interactions for specific call types.
⚙️ Applied redaction to pipeline LLM resources: Configured all pipeline-related LLM, embedding, and table refinement resources to automatically send message redaction headers, ensuring efficient log management for data processing workflows.

Fixed

🐛 Resolved LiteLLM user service compatibility: Adjusted user key retrieval logic to correctly handle HTTP 404 responses from the updated LiteLLM proxy when a user does not exist, ensuring robust user management.

[v0.284.0] - 2026-04-28 - Introducing Continuous Postgres Maintenance for Dagster

Added

✨ Automated Dagster Database Maintenance: Launched a comprehensive subsystem within the existing backup Dagster instance to keep the dagster database bounded and healthy over long-running deployments. This includes:
- 🧹 Weekly Log Cleanup: A new dagster_cleanup_job that automatically prunes verbose Python logs (DEBUG, INFO, WARNING) and transient framework-internal events (e.g., HANDLED_OUTPUT, LOADED_INPUT, ENGINE_EVENT) from the event_logs table, based on configurable retention policies.
- 📦 Monthly Disk Space Reclamation: A postgres_repack_job that runs pg_repack on heavy tables like event_logs, runs, and job_ticks to physically return unused disk pages to the operating system.
- 🛡️ UI-Safe Pruning Guarantees: The cleanup process is meticulously designed to never delete critical data required by the Dagster UI, such as asset materializations, step success/failure records, run history, or asset catalog entries.
- ⚙️ Configurable Retention & Batch Limits: New environment variables (DAGSTER_DEBUG_LOG_RETENTION_DAYS, DAGSTER_INFO_LOG_RETENTION_DAYS, DAGSTER_WARNING_LOG_RETENTION_DAYS, DAGSTER_UNIMPORTANT_EVENT_RETENTION_DAYS, DAGSTER_CLEANUP_BATCH_LIMIT, MAINTENANCE_DISABLED) provide fine-grained control over retention windows, cleanup performance, and a kill switch for maintenance.
🐳 Custom Postgres Image with pg_repack: Integrated a project-managed Postgres Docker image (pgvector-repack:pg17) that extends the base pgvector image with the postgresql-17-repack extension, a prerequisite for efficient disk space reclamation.
✅ Extensive Testing for Database Maintenance: Added new layers of unit and integration tests, including SQL contract tests against a real Postgres instance, to ensure the correctness and safety of the new maintenance subsystem.

Changed

🔄 Serialized Postgres Operations: Implemented tag-based concurrency limits using Dagster's QueuedRunCoordinator for all Postgres-affecting jobs (backup, restore, cleanup, and repack). This ensures these critical operations run sequentially, preventing conflicts and improving system reliability.
📄 Enhanced Documentation for Data Retention & Maintenance: Updated architecture decision records (arc42) and user guides with detailed explanations of the new continuous Postgres maintenance subsystem, covering its purpose, configuration, safety guarantees, and operational implications.
🕰️ Dagster Tick History Retention: Configured automatic retention policies for Dagster job and sensor tick history in both the main and backup Dagster instances, preventing unbounded growth of internal tick metadata.
⚙️ Streamlined Image Management: Updated internal infrastructure configurations and Makefile targets for building and publishing project-managed Docker images (Postgres and Playwright), allowing for local builds in development and consistent pulls in other stages.

[v0.283.0] - 2026-04-28 - Smarter Image Handling with Perceptual Deduplication

Added

✨ Introduced Perceptual Deduplication for Images: Implemented new logic to perceptually hash images during document processing, enabling the identification and reuse of identical or near-identical figures (e.g., logos) to significantly reduce storage and upload overhead.
🧮 Enabled Content-Addressed Filenaming: Uploaded image files are now named using a SHA256 content hash, which ensures idempotency and automatically collapses bytewise duplicate images.
🖼️ Enhanced Image Format Detection: Improved the mechanism for determining appropriate image file extensions, leveraging data URI MIME types, original filenames, and robust PIL sniffing.
🧪 Added Comprehensive Image Processing Tests: A new dedicated test suite ensures the reliability and correctness of image extraction, hashing, and upload functionalities.

Changed

🚀 Optimized Image Upload Workflow: The extract_and_upload_images utility now intelligently processes images, checking for perceptual matches against previously uploaded figures before initiating new S3 uploads.
📄 Standardized Markdown Figure Tags: Markdown image references are now consistently wrapped in <figure> tags after being uploaded to S3, improving semantic structure.

[v0.282.0] - 2026-04-28 - Granular RAG Execution Outcomes for Better Visibility and Control

Added

✨ Introduced Granular RAG Stop Events: New event types (RAGSuccessStopEvent, RAGFailureStopEvent, and RAGStopEvent as their common ancestor) and an associated RAGFailureReason enumeration have been added. These provide detailed and specific outcomes for Retrieval-Augmented Generation (RAG) agent runs, distinguishing between successful responses and various failure conditions.
🦾 Implemented do_finalize_rag_stop Logic: A new step function, do_finalize_rag_stop, was introduced to intelligently determine the final RAG outcome. It consolidates information from the LLM's response and any prior rejection events (such as few-shot rejection, insufficient context, or expert decline/error) to emit the most appropriate RAG stop event.
🌐 Integrated RAG Outcome Display in UI: Dedicated UI components and internationalized messages have been added to visually represent RAG success and various failure events within the application, offering clearer, contextualized feedback to users.

Changed

🔄 Updated RAG Agent Termination Flow: Both the RAG and Expert RAG agents now utilize the new RAGSuccessStopEvent or RAGFailureStopEvent for run termination, replacing the generic StopEvent. This change provides richer, more contextualized information about the completion status of agent runs.
💬 Improved Expert RAG Agent Error Handling: The Expert RAG agent's internal response paths for when an expert declines to answer or encounters an error now explicitly emit a RAGFailureStopEvent with a specific reason, enhancing the clarity of situations where a grounded answer could not be produced.
🛠️ Refined Context Sufficiency Guard Logic: The do_context_sufficient_guard step has been updated to correctly emit a ContextInsufficientRejectEvent when the maximum number of retrieval hops is exhausted, ensuring accurate reporting for ungrounded answers due to insufficient context.

[v0.281.1] - 2026-04-28 - Routine Version Increment

Changed

🔄 Version alignment: Updated the main project and all associated packages (agent, api, backup, bot, core, pipeline, process, web) to version 0.281.1.

[v0.281.0] - 2026-04-27 - Enhanced Dagster Data Lifecycle Management

Added

✨ Introduced Dedicated dagster Bucket: A new S3 bucket named dagster is now automatically provisioned during infrastructure setup, specifically for storing Dagster's intermediate operational data.
⏳ Automated Expiration for Intermediate Data: The newly created dagster bucket is configured with a 1-day time-to-live (TTL) policy, ensuring automatic cleanup of temporary pipeline execution data to optimize storage.

Changed

🔄 Optimized Dagster Intermediate Storage: The S3PickleIOManager used by default_io_manager_s3_datalake_resources now directs all intermediate pipeline data to the dedicated dagster bucket, utilizing a streamlined container_name/ prefix for improved organization and adherence to the new lifecycle management.
📄 Updated Resource Documentation: Documentation for default_io_manager_s3_datalake_resources has been updated to clarify that intermediate data is now stored in the dagster bucket with a 1-day expiration policy, providing accurate information on data persistence.

Refactor

🧹 Improved Intermediate I/O Manager Naming: The internal variable for the S3PickleIOManager was renamed to op_intermediates_io_manager for enhanced clarity and better reflection of its purpose within the pipeline resources.

[v0.280.1] - 2026-04-27 - Precision Testing: Unit/Integration Split & Keycloak Validation

Added

✨ Enhanced Pytest Utilities: Introduced new generic utilities (mark_tests_by_directory, attach_fixtures_to_items) to automatically categorize tests as unit or integration and explicitly manage fixture application, significantly improving test suite organization.
🔑 Robust Keycloak Integration Tests: Implemented comprehensive integration tests for the KeycloakAdminService, covering user management, tenant group operations, and active tenant attribute handling to ensure reliable interaction with Keycloak.
🛠️ Real Keycloak Admin Client Utility: Provided a new utility function to easily create and configure real Keycloak Admin clients for dedicated use in integration testing environments.

Changed

🚀 Expanded CI/CD Test Environments: Updated GitHub Actions workflows to include a Keycloak service in the test environments for the core and api packages, enabling more thorough integration testing.
⚡️ Improved Test Execution Commands: Introduced granular Makefile targets (test, test-integration, test-all) in the backup and core packages, allowing developers to run specific sets of unit or integration tests.
🏷️ Standardized Pytest Markers: Defined official unit and integration pytest markers in pyproject.toml files for the backup and core packages, replacing previous conventions for clearer test categorization.

Refactor

🧹 Comprehensive Test Suite Restructuring: Performed a significant overhaul of the entire test suite, relocating numerous test files across agent, api, backup, bot, core, pipeline, and process packages into dedicated unit/ and integration/ subdirectories for clearer separation and management.
🔄 Unified Fixture Management: Transitioned many fixtures (including database isolation, Keycloak mocks, tenant mocks, role mocks, MSAL/aiohttp mocks) from implicit autouse=True to explicit attachment via pytest_collection_modifyitems hooks and pytestmark in conftest.py files, offering more granular control over test setup.
📄 Minor Configuration Description Clarity: Adjusted the prompt description for context_insufficient_prompt in the agent's ContextSufficientGuardStepConfig for improved readability.

[v0.280.0] - 2026-04-27 - Proactive Pipeline Monitoring with Apprise Notifications

Added

✨ Introduced Pipeline Run-Failure Notifications: A new, comprehensive system leveraging dagster-apprise to provide immediate alerts for failed pipeline runs across all code locations. This includes failures in both explicit jobs and automatically materialized assets, ensuring better visibility into pipeline health.
⚙️ Configurable Notification Settings: Added new environment variables (NOTIFICATION_URLS, NOTIFICATION_DAGSTER_UI_BASE_URL, NOTIFICATION_TITLE_PREFIX, NOTIFICATION_MIN_INTERVAL_SECONDS) to easily configure diverse notification channels (e.g., Slack, Teams, email, PagerDuty) and customize notification behavior.
📚 Integrated NotificationSettings Module: A new Pydantic-based settings module was added to centralize and robustly parse, validate, and manage all notification-related environment variables.
🔗 Automatic Sensor Wiring for Default Pipelines: Default pipeline definitions now automatically include the run-failure notification sensor when configured via environment variables, ensuring broad coverage without requiring manual code changes for standard deployments.
📄 New Architectural Decision Record (ADR): A detailed ADR documenting the rationale, design choices, and trade-offs of the new pipeline run-failure notification system has been added to the documentation.
📖 Updated Pipeline Documentation: The pipeline documentation (CLAUDE.md) has been updated with a dedicated section explaining the new run-failure notification capabilities, including configuration and usage.
📦 Added dagster-apprise and apprise Dependencies: Integrated the necessary libraries to power the multi-channel notification system.

Changed

🔄 Deployment Configuration Updates: Docker Compose templates and generated files have been updated to dynamically provision the new notification-related environment variables to pipeline services, streamlining deployment setup for the new feature.

Refactor

🧹 Improved Makefile Declarations: Added .PHONY declarations to the pipeline Makefile, enhancing clarity and reliability for build and test commands.

Changed

✨ Improved Organization Memory Accessibility: RAG agents and the core AgentMemory service now default to tenant-wide retrieval for organization memory by explicitly passing user_id=None and agent_id=None to the underlying search function. This change ensures that knowledge stored in organization memory is fully shared and broadly accessible across all agents within a tenant, promoting consistent information access.
📄 Expanded Memory Scoping Documentation: Updated the AgentMemory service documentation with detailed explanations on how organization memory is shared across agents and how to control user_id scoping for either fully shared tenant-wide or user-specific retrieval.

Added

🧠 New Shared Knowledge Agents: Introduced Shared Knowledge Selector and Shared Knowledge RAG agent templates, enabling agents to intelligently route and retrieve information from designated shared knowledge buckets.
🛡️ Dedicated Context Sufficiency Guard Configuration: A new ContextSufficientGuardStepConfig allows for granular control over how agents verify if retrieved context is adequate to answer a user's query, improving response accuracy.
💬 Chat History Formatting Utility: Added format_chat_history to consistently present conversation history for LLM prompts, enhancing the context provided to AI models.

Changed

⚙️ Refined RAG Agent Configuration: The RAG and Expert RAG agent configurations now embed context sufficiency guard settings within the new ContextSufficientGuardStepConfig, streamlining agent setup.
🔍 Enhanced Context Guard Logic: The core context sufficiency guard now utilizes the full chat history, including organizational memory, to make more informed decisions about whether sufficient information is available.
📝 Updated Context Guard Prompts: Adjusted the internal prompts for the context sufficiency guard across all supported languages to incorporate chat history for better decision-making.

Fixed

🐛 Improved UI Form Hydration: Resolved an issue where agent and process creation forms failed to hydrate correctly when nested optional configurations were explicitly null, enhancing the reliability of agent management in the UI.
📦 Robust Human-in-the-Loop Dispatch: Enhanced the chat service's ability to correctly identify and dispatch appropriate Human-in-the-Loop (HITL) response events based on the type of HITL request, preventing potential interaction errors.

Refactor

🔄 Standardized Agent Template Loading: Refactored how templates for Few-Shot and LLM-Wrapping agents are loaded, moving from static TEMPLATE variables to dynamic build() functions for greater flexibility and maintainability.
🧹 Centralized Context Sufficiency Settings: Consolidated all configurations related to context sufficiency checks into the new ContextSufficientGuardStepConfig, simplifying agent codebases and improving modularity.

[v0.279.0] - 2026-04-24 - Advanced RAG Filtering and Configuration Management

Added

✨ Introduced Per-Run RAG Metadata Filtering: Publishers can now supply additional_filters with RAGStartEvent to dynamically narrow retrieval results based on metadata keys, allowing for more granular control over RAG contexts.
🦾 New RetrievalRuntimeConfig Model: A new plain Pydantic BaseModel (RetrievalRuntimeConfig) has been added to cleanly separate design-time agent configuration from per-run runtime overrides. This enhances modularity and security by preventing publisher-only fields from appearing in the Admin UI.
⚙️ Configurable Allowed Metadata Filter Fields: Agent administrators can now define a whitelist of metadata keys (allowed_metadata_filter_fields) within the MilvusVectorStoreConfig for each retriever, ensuring publishers can only filter on explicitly permitted fields.
🌐 Admin UI for Allowed Filter Fields: The Admin UI's VectorStoreInput now includes a dedicated control for specifying allowed_metadata_filter_fields, making this powerful new RAG filtering capability easy to configure.
📊 Display of RAG Start Event Filters: The agent event stream now visually presents additional_filters and selected_namespaces when a RAGStartEvent occurs, providing better visibility into retrieval parameters.
📄 Architectural Decision Record: A new architectural decision record (2026_04_22_form_runtime_config_wrapper.md) has been added, documenting the rationale and pattern for separating design-time configurations from runtime overrides.

Changed

🔄 Updated RAG Retrieval Logic: The internal RAG retrieval steps now leverage the new RetrievalRuntimeConfig wrapper, allowing for the application of per-run additional_metadata_filters directly during the node retrieval process.
🚀 Optimized API Event Payloads: StartEvent input data in agent discovery services now uses exclude_none=True when dumping models to JSON, resulting in smaller and more efficient wire payloads.
📚 Refined Event Protocol Documentation: Examples in the Swiss AI Agent Protocol documentation were updated to reflect the use of message blocks instead of plain content for UserMessageEvents.

Refactor

📦 Relocated RAGStartEvent: The RAGStartEvent has been moved to the core package (swiss_ai_hub.core.events.agent.control.start) to make it more broadly accessible across the application.
🧹 Replaced Namespace Filtering Utility: The filter_retrievers_by_namespace utility has been replaced by the more comprehensive narrow_retrievers function, which now handles both namespace narrowing and additional_metadata_filters validation.

Removed

🗑️ Deprecated filter_retrievers_by_namespace: The previous utility for filtering retrievers by namespace has been removed, as its functionality is superseded by the new narrow_retrievers mechanism.

[v0.278.1] - 2026-04-23 - Strengthened Event Serialization for HITL Subtypes

Fixed

🐛 Resolved Human-in-the-Loop (HITL) Event Downcasting: Fixed a critical bug where specific HITL event subclasses (such as input, confirmation, and chat requests and their corresponding responses) were silently downcasted to their parent event types during WebSocket serialization. This ensures that the frontend now receives accurate and specific HITL event types.
⚡️ Corrected Event Discriminator Logic: Enhanced the DisplayEvents union to correctly discriminate between agent-specific HITL request and response event subclasses, preserving their distinct type information when events are broadcast over WebSockets.

Added

✅ Comprehensive Event Serialization Tests: Introduced new unit tests to rigorously validate that Human-in-the-Loop event subclasses are properly discriminated and their precise type is preserved throughout the WebSocket serialization and deserialization process, preventing future regressions.

Changed

📄 Updated Event Serialization Guidelines: Added crucial documentation to CLAUDE.md files in both api and core packages, providing clear guidance on how to correctly add new event subclasses to the DisplayEvents union to prevent silent downcasting during WebSocket serialization.

[v0.278.0] - 2026-04-21 - Next-Gen Multi-Tenancy: Keycloak-First Identity & Admin Workflows

Added

🦾 Dedicated Sysadmin Tenant Management API: Introduced a new set of API endpoints (/admin/tenants) for system administrators to manage tenant metadata, including listing, configuring, updating, and deleting tenants.
🔑 AIHubSysAdmin Keycloak Realm Role: A new Keycloak realm role, AIHubSysAdmin, now designates platform administrators, granting implicit admin-level access within any tenant they are a member of.
🆕 TenantMetadataEntity for Tenant Display Information: A new database entity TenantMetadataEntity is introduced to store display-related tenant information (name, description, access rules), complementing Keycloak as the authoritative source for tenant existence.
🧪 TestAuthHandler for Development & Testing: A dedicated authentication handler (TestAuthHandler) replaces the DangerousDevelopmentOnlyAuthHandler, ensuring a clearer separation between test utilities and production code and simplifying local development setup.
🔄 Automatic Superuser Membership in New Tenants: The platform's superuser account (now a regular Keycloak user) is automatically added to every new tenant's Keycloak group upon creation, ensuring immediate and explicit access for platform administration.
📄 New Architectural Decision Records (ADRs): Multiple new ADRs have been added to thoroughly document the rationale and implementation details of significant architectural changes in multi-tenancy, authorization, and token management.

Changed

🔑 Superuser Identity Transition: The global superuser is no longer a synthetic identity but a regular Keycloak user, identified by SUPERUSER_USERNAME and SUPERUSER_EMAIL, simplifying identity management and leveraging Keycloak as the single source of truth.
⚙️ SUPERUSER_TOKEN Repurposed: The SUPERUSER_TOKEN now serves as a static bearer token, bound to the Keycloak superuser, enabling internal services to authenticate against the API through the standard TokenAuthHandler.
🚀 StartupTenantSettings Introduced: The previous AIHUB_DEFAULT_TENANT_ID and related settings are now AIHUB_STARTUP_TENANT_ID and StartupTenantSettings, clarifying that this tenant is treated as an ordinary tenant after initial seeding, losing its special "default" status.
⚡️ Refined AuthHandler Tenant Resolution: The logic for resolving active and requested tenants has been updated, with KeycloakAdminService now the authoritative source for tenant existence and membership checks, improving consistency and reliability.
👤 User Profile Data Sourced from Keycloak: User profile data (name, email) is now exclusively read from Keycloak (JWT claims for OAuth2, KeycloakAdminService for bearer tokens), eliminating the need for a local UserEntity in MongoDB.
📚 Updated Documentation for Admin/Sysadmin UI: User-facing documentation and UI layouts for admin functionality are now separated, with dedicated sections for tenant-scoped admin pages (/[tenant]/service/) and global sysadmin pages (/sysadmin/).
🔗 Dynamic Endpoints Include Tenant ID: Dynamic Agent and Process endpoints now explicitly include the tenant ID in their URL paths: POST /api/v1/{tenant_id}/agents/{agent_class}/{agent_id}/{event_name}.
📈 get_my_tenants Returns Sysadmin Status: The /my-tenants API endpoint now includes the user's is_sys_admin status in its response, providing a clearer overview of user privileges.
🛡️ Bot API Authentication Hardened: The Bot API's main entry point (packages/bot/app/main.py) now uses KeycloakAuthHandler as a fail-closed safety net, replacing the DangerousDevelopmentOnlyAuthHandler placeholder.
📊 Stricter MONGO_MAIN_DB_NAME Validation: Updated validation for AIHubSettings().MONGO_MAIN_DB_NAME to ensure stricter adherence to valid MongoDB database naming conventions.
🛠️ Agent RunContext/ThreadContext Backing Changed: RunContext and ThreadContext for agents are now backed by Valkey (Redis), providing ephemeral (RunContext) and persistent (ThreadContext) key-value storage.
🔗 Keycloak Service Account Permissions Extended: The aihub-api-service Keycloak client now requires view-realm and view-clients roles in realm-management to enable proper sysadmin status resolution.

Fixed

🐛 PostgreSQL FerretDB Catalog Backup Workaround: Implemented a specific workaround for PostgreSQL FerretDB catalog tables to ensure their data is correctly backed up and restored, preventing empty catalogs after recovery.
🐛 MongoDB Test Database Isolation: Ensured robust isolation of MongoDB test databases by enforcing an early import in conftest.py that sets AIHUB_MONGO_MAIN_DB_NAME to aihub_test, preventing accidental data leakage to dev/prod databases.
🐛 Token Verification Error Messages: Improved error messages for bearer token verification to distinguish between malformed, unknown, and expired tokens, aiding in debugging and troubleshooting.

Removed

🗑️ TenantEntity Database Collection: The TenantEntity MongoDB collection has been completely removed, with its metadata functionality replaced by TenantMetadataEntity and existence managed by Keycloak.
❌ DangerousDevelopmentOnlyAuthHandler: The insecure DangerousDevelopmentOnlyAuthHandler has been entirely removed, promoting more secure authentication practices even in development environments.
🚫 SuperuserAuthHandler: The dedicated SuperuserAuthHandler has been removed as the superuser is now an ordinary Keycloak user authenticated via the standard TokenAuthHandler.
🗑️ System-Wide Roles: The concept and implementation of system-wide roles (roles with tenant_id=None) have been removed; all roles are now strictly tenant-scoped.
🗑️ is_default Flag on Tenants: The is_default boolean flag has been removed from tenant metadata, simplifying the tenant model and delegating "startup tenant" identification to configuration.
🗑️ Legacy KEYCLOAK_DEV_USER_* Environment Variables: The KEYCLOAK_DEV_USER_* environment variables, previously used for a configurable development user, have been removed in favor of the more robust SUPERUSER_* configuration.

Security

🔒 Opaque API Bearer Token Format: API bearer tokens now use an opaque sk-<random> format, enhancing security by removing sensitive database internal IDs and improving token opacity, requiring rotation of existing tokens.
🛡️ Sysadmin Authorization Bypass Refinement: Sysadmin access now grants ACCESS_ADMIN within any tenant the principal is a member of, bypassing normal rule evaluation, while explicitly NOT granting implicit membership, aligning with the principle of least privilege.
👮 Stricter Tenant Request Validation: New DTOs and validation rules for tenant administration API endpoints prevent malformed or invalid tenant IDs and names from reaching business logic or Keycloak, mitigating potential injection risks.
🚨 Explicit Auth Handler Requirement: Controller.__init__(auth=...) now explicitly raises a TypeError if no auth handler is provided, preventing silent activation of insecure bypass mechanisms.
🕵️ Tenant Enumeration Protection: Tenant resolution via URL path parameter now consistently returns a generic "Access denied" (403) for non-existent, non-member, or unconfigured tenants, preventing attackers from enumerating valid tenant IDs.
🛑 Last Tenant Deletion Guard: The tenant deletion service now explicitly prevents the removal of the last remaining tenant, ensuring continuous platform availability and integrity.

Refactor

🧹 Consolidated Superuser Configuration: The configuration for the platform superuser is consolidated into a single SUPERUSER_ block in .env files, simplifying setup and maintenance.
🧽 Race-Safe Tenant Deletion Logic: The tenant deletion process is refactored with robust race-safety mechanisms, including snapshotting and a post-deletion recount, to ensure the "at least one tenant must remain" invariant is never violated by concurrent operations.
🧹 Centralized Keycloak Admin Service: The KeycloakAdminService is significantly expanded with new static methods for comprehensive tenant group, user membership, and realm role management, centralizing Keycloak interactions and improving testability.
🧹 Streamlined Auth Handler Composition: The TokenAndOauth2Handler composition logic is streamlined, removing unused authentication paths and clarifying the overall authentication flow.
🧹 Unified Test Identity Management: Test identity constants (OID, name, email, roles) and a fake_user() helper are moved to swiss_ai_hub.core.testing.auth_utils.test_identity.py, explicitly separating test-only values from production settings.
🧹 Updated AccessChangeHook for New Entities: The AccessChangeHook now correctly tracks changes on TenantMetadataEntity instead of the removed TenantEntity, maintaining consistency for OpenWebUI synchronization.
🧹 Standardized Agent Playground Triggers: Agent playground trigger scripts now consistently use fake_user() from the new core.testing.auth_utils for mock user identities, promoting uniformity and ease of testing.
🧹 Consolidated Role Card Deletion Logic: Role card deletion in the UI now uses a unified confirmation dialog and delegates to a shared composable for improved user experience and maintainability.
🎨 Standardized UI Loading Indicator: Introduced a new AppLoader component for a consistent and visually appealing loading indicator across the UI.
🧹 General Code Cleanup and Consistency: Various minor code cleanups, including improved logging, removal of redundant code, and consistency adjustments across the codebase.

[v0.277.2] - 2026-04-20 - Refined Event Protocol: Introducing `RAGStartEvent` and Clarifying Chat Contracts

Added

✨ Introduced RAGStartEvent: A new, dedicated StartEvent for RAG agents. This event allows custom domain front-ends or other agents to initiate RAG workflows with pre-selected namespaces, maintaining a clean separation from the generic chat UI contract.
🧪 Unit Tests for RAGStartEvent: Comprehensive unit tests were added to validate the properties and behavior of the new RAGStartEvent, ensuring its reliability.
🖼️ Web UI Display for RAGStartEvent: A new Vue component and updated event mapping were implemented to enable proper rendering of RAGStartEvent in the web interface, showcasing selected namespaces and chat history.
🌐 Internationalization for RAGStartEvent: Added translations across supported languages for the RAGStartEvent and its associated "selected namespaces" property, enhancing multi-language support.

Changed

📄 Clarified UserMessageEvent as Chat UI Contract: Updated documentation and internal guidance across SKILL.md, agent reports, and CLAUDE.md files. This change emphasizes that UserMessageEvent is exclusively for generic chat interfaces and advises subclassing StartEvent directly for agent- or domain-specific payloads.
🔄 RAG Agent Event Acceptance: Modified ExpertRAGAgent and RAGAgent to explicitly accept the new RAGStartEvent (alongside UserMessageEvent) for initiating RAG workflows, ensuring proper handling of namespace-aware queries from non-chat sources.

Refactor

🧹 Replaced NamespaceAwareUserMessageEvent with RAGStartEvent: Refactored internal agent logic and configurations to transition from NamespaceAwareUserMessageEvent to the newly introduced RAGStartEvent, streamlining the event model for RAG delegation.
⚡️ Optimized UserMessageEvent.user_query Property: Performed a minor internal optimization to the user_query property within UserMessageEvent for enhanced readability and maintainability.

Removed

🗑️ Deprecated NamespaceAwareUserMessageEvent: The NamespaceAwareUserMessageEvent class was removed, as its functionality has been superseded by the more appropriately designed RAGStartEvent which directly extends StartEvent.

[v0.277.1] - 2026-04-17 - Enhanced Development Experience and Service Orchestration

Added

✨ New IntelliJ Run Configurations: Introduced dedicated IntelliJ run configurations for Bot Dev and Bot Prod to streamline local development and testing of bot services using Makefiles.

Changed

🚀 Improved IntelliJ Docker Compose Configurations: Enhanced existing Docker Compose run configurations in IntelliJ by explicitly defining environment file paths and ensuring services are rebuilt on each run.
⚙️ Streamlined Open WebUI Dependencies: Optimized Docker Compose configurations to remove the direct dependency of open-webui on the api service's health in non-development environments, improving startup times.
♻️ Enhanced openwebui-init Resilience: Updated the openwebui-init service in Docker Compose to automatically restart on failure, improving the stability and robustness of initialization tasks.
⏱️ Ensured Sequential Service Startup: Configured openwebui-manager to wait for openwebui-init to successfully complete its tasks before starting, preventing potential race conditions and ensuring correct setup.

Removed

🗑️ Deprecated Bot Run Configuration: Removed the outdated IntelliJ Python run configuration for "Run Bots," now superseded by the new, more specific Makefile-based configurations.

Refactor

🧹 Centralized Keycloak Client Dependency: Relocated the python-keycloak dependency from the packages/api module to packages/core, centralizing common authentication functionalities.

[v0.277.0] - 2026-04-14 - Enhanced Data Resilience: Introducing a Centralized Backup and Restore Service

Added

🚀 New Centralized Backup and Restore Service: Introduced a comprehensive, Dagster-powered backup service that ensures data integrity across all stateful services, including PostgreSQL (main and FerretDB), Milvus, Neo4j, ClickHouse, Valkey, and NATS.
⏰ Automated Daily Backups: Configured daily backup schedules at 1 AM (Europe/Zurich) with automatic retention cleanup policies, managed via the dedicated Dagster UI.
↩️ Self-Service Restore Capabilities: Enabled parameterized point-in-time recovery for the entire platform through the intuitive Dagster Launchpad in the Backup UI.
📄 Extensive Backup Documentation: Provided detailed documentation covering the backup architecture, configuration, operational procedures, and recovery steps for all data stores.
🧪 End-to-End Backup Tests: Implemented a robust end-to-end (E2E) test suite to validate the entire backup and restore lifecycle, ensuring reliability and data integrity in an integrated environment.
📦 Dedicated Backup S3 Bucket: Automatically provisioned a new backups bucket within the internal SeaweedFS S3 storage for all backup artifacts.
🛠️ Native Backup Tooling Integration: Incorporated official milvus-backup and nats CLI tools directly into the backup service for precise and efficient data handling.
⚙️ ClickHouse S3 Integration: Configured ClickHouse to directly utilize the internal SeaweedFS S3 for its native BACKUP TO Disk() and RESTORE FROM Disk() operations.
📦 New packages/backup Module: Introduced a new top-level Python package to house all the logic for the centralized backup and restore service.

Changed

🗃️ Improved Backup Retention Logic: Enhanced the backup retention policy to guarantee a minimum number of backups are always preserved, even if they exceed the defined age limit.
📈 Milvus Collection Loading Reliability: Updated Milvus vector store operations to explicitly ensure collections are loaded into memory before use, preventing potential issues after service restarts or restores.
🏗️ Docker Compose Integration: Updated core Docker Compose configurations across all stages to seamlessly integrate the new backup service, manage its dependencies, and define necessary environment variables.

Fixed

🐛 DocumentDB Catalog Backup Fix: Resolved a critical issue where pg_dump would silently skip essential DocumentDB catalog tables in FerretDB's PostgreSQL backend. The backup service now explicitly backs up and restores these tables and sequences using the COPY protocol, ensuring consistent FerretDB restores.

Removed

🗑️ Deprecated Backup Stubs: Removed placeholder (stub) implementations for backup handlers, replacing them with fully functional and robust service-specific backup logic.

[v0.276.0] - 2026-04-14 - Streamlined Workflow Visualization and Enhanced Multi-Tenancy

Refactor

🧹 Agent Workflow Visualizer: Completely re-engineered the backend workflow visualizer, eliminating the networkx dependency and streamlining the internal graph representation for improved efficiency.
🔄 Frontend Workflow Visualization: Overhauled the frontend workflow visualization to use a custom grid-based layout algorithm, providing a more intuitive and performant display of agent structures without relying on external graphing libraries.
🦾 Simplified Workflow Graph Models: Refactored the underlying data models (WorkflowGraph, NodeData, EdgeData) for agent workflows, removing verbose event details from the graph structure for a cleaner and more focused representation.
🏗️ Standardized Workflow Node Components: Restructured workflow node components (StartNode, StepNode, StopNode) to utilize a new WorkflowNodeCard for consistent styling and updated connection handle positions for better layout integration.
⚡️ Improved useTenantPath hook usage: Applied minor refactorings to ensure consistent and efficient usage of the useTenantPath hook across the application.

Added

🖼️ Agent Workflow Modal: Introduced a new modal component for a dedicated, expanded view of agent workflow diagrams, improving user experience for understanding agent logic.
✨ Generic Workflow Node Card: Added a reusable WorkflowNodeCard component to standardize the appearance of nodes across the new workflow visualization, ensuring a consistent look and feel.
🚀 View Workflow Button: Implemented a "View workflow" button on the agent class list, allowing quick access to the visual representation of an agent's logic directly from the overview.
📄 Localization for Workflow UI: Added new localization strings to support the user interface elements of the enhanced workflow visualization in multiple languages.

Changed

🔑 Enforced Tenant ID in API Calls: Updated numerous frontend API calls to explicitly include the tenantId, strengthening multi-tenancy enforcement and data isolation across various services.
📊 Streamlined WorkflowGraph Data Model: Simplified the WorkflowGraph data model by removing directed, multigraph, and generic graph attributes, focusing solely on nodes and links for a leaner representation.

Removed

🗑️ Dropped Graphing Library Dependencies: Eliminated networkx from backend dependencies and @dagrejs/dagre from frontend dependencies, significantly reducing the package footprint and improving performance.
🚫 Deprecated Detailed Event Models: Removed Pydantic and MongoEngine models (EventInfo, EventPayloadField, InputEventInfo) that previously stored extensive event details within the graph, aligning with the simplified visualization approach.
🧹 Legacy Workflow UI Components: Removed old Vue components (WorkflowEventEdge.vue, WorkflowEventSpecs.vue) that are no longer needed due to the redesigned workflow visualization.

[v0.275.0] - 2026-04-14 - Keycloak as Source of Truth: Centralized User Management and Active Tenant Handling

Refactor

🔄 Centralized User Management: Rearchitected user management by migrating user profile data (name, email) and active tenant ID from the local MongoDB UserEntity collection to Keycloak. Keycloak now serves as the single source of truth for user identities, reducing data duplication.
🧹 User Dashboard Persistence: Refactored the storage of user dashboard configurations into a new, dedicated MongoDB collection, UserDashboardEntity, separating it from core user identity data.
⚡️ Enhanced Keycloak Admin Permissions: Updated the aihub-api-service Keycloak client with necessary manage-users and query-users roles, enabling the application to perform user and group management operations directly via the Keycloak Admin API.
⚙️ Configurable Default Tenant ID: Standardized the default tenant ID (AIHUB_DEFAULT_TENANT_ID) across environment variables and Keycloak realm configurations for improved consistency and flexibility.
📄 Internal API and Bot Codebase: Performed extensive updates across the API and bot services to align with the new user management architecture, removing all direct dependencies on the defunct UserEntity and migrating to asynchronous Keycloak Admin Service calls.

Added

🔑 Keycloak Admin Service: Introduced a new, robust KeycloakAdminService with methods for user lookup, creation, and management, including reading and writing custom user attributes (like active_tenant_id) and managing tenant groups.
📊 Dedicated User Dashboard Entity: Implemented UserDashboardEntity to manage user dashboard configurations, replacing the dashboard functionality previously embedded in UserEntity.
✨ Keycloak Data Models: Introduced KeycloakUser and KeycloakGroup Pydantic models for structured representation of Keycloak entities, improving type safety and development experience.
🚀 Automated Keycloak Tenant Group Provisioning: Added logic to automatically create corresponding Keycloak tenant groups for default tenants and to backfill existing Keycloak users into these groups upon initial setup.

Removed

🗑️ Elimination of UserEntity: Completely removed the UserEntity MongoDB collection, simplifying the persistence layer by centralizing user identity in Keycloak.
⛔️ Removed User-Local Metadata: Discontinued local storage of last_accessed timestamps and favorite_modules lists from user profiles, reflecting the shift to Keycloak as the primary user data store.
🚫 Superuser Database Record: The superuser no longer creates a persistent record in the local database, operating purely as a virtual identity for system administration.

[v0.274.4] - 2026-04-14 - Introducing Fine-Grained OpenWebUI Agent Visibility & Workflow Enhancements

Added

🛡️ Implemented OpenWebUI Agent Visibility Provisioning: AI-Hub now actively manages OpenWebUI's internal state, including groups, workspace models, and access grants. This ensures users only see agents they are authorized to use, enforcing granular access control based on AI-Hub's permission system.
✨ New Webhook Controller: Introduced a WebhookController in the API to receive and process external service notifications, currently used for OpenWebUI user signup events to trigger access synchronization.
⚙️ Access Change Hook: Added an AccessChangeHook to automatically trigger OpenWebUI access synchronization when AI-Hub roles, tenants, or user-role assignments are created, updated, or deleted, with built-in debouncing for performance.
📄 OpenWebUI Access Control Documentation: New architecture decision records and user documentation have been added to explain the new OpenWebUI model visibility and access control mechanisms.
🧪 Comprehensive OpenWebUI Provisioner Tests: Expanded unit and integration tests for OpenWebUI provisioning, covering group management, access grant computation, and distributed locking.
🛠️ Utility Methods for Agent Discovery: Introduced AgentClassEntity.get_online_conversational() and AgentConfigEntityDocument.find_for_classes() to efficiently retrieve online conversational agent data for provisioning.

Changed

🚀 Upgraded OpenWebUI Dependency: Updated OpenWebUI Docker image to v0.8.10, enabling new features and security enhancements required for integrated access control.
🎯 Refined PR Feedback Skill: The implement-feedback-from-pr skill now intelligently fetches only unresolved, non-outdated review comments from GitHub, streamlining the feedback implementation process.
🎥 PR Demo Video Workflow: The pr-demo-video skill now posts demo video descriptions as comments directly on the linked GitHub issue, falling back to the PR if no issue is connected, improving PR management and clarity.
🔑 OpenWebUI Configuration Parameters: Added new environment variables (OPENWEBUI_SCIM_TOKEN, OPENWEBUI_WEBHOOK_SECRET) and updated Docker Compose configurations to enable OpenWebUI's SCIM and webhook capabilities, alongside disabling model access control bypass.
🔄 API Lifetime Management: Integrated the OpenWebUI provisioner into the API's lifecycle manager, ensuring access control is applied automatically on startup and updated when access entities change.
⚡️ Optimized Agent Instance Hashing: Improved the internal logic for computing agent instance hashes, enhancing the efficiency of change detection for provisioning services.
📧 Keycloak Development User Email: Updated the default Keycloak development user email domain for consistency.
👥 User Active Tenant Sync Trigger: Modified the UserEntity.set_active_tenant method to explicitly notify the AccessChangeHook, ensuring OpenWebUI permissions are re-evaluated when a user changes their active tenant.

Refactor

🧹 OpenWebUI Initialization Script Renaming: Renamed infra/configs/openwebui/init-functions.sh to init-openwebui.sh for improved clarity and consistency with its expanded role in initializing OpenWebUI functions and service accounts.
⚙️ Internal Provisioner Dependencies: Refactored the AgentEndpointsDiscoveryService to internally instantiate LangfuseProvisioner and OpenWebuiProvisioner, simplifying its constructor and centralizing dependency management.

[v0.274.3] - 2026-04-10 - Addressing Duplicate Agent-in-the-Loop Responses and Enhancing Event Control

Fixed

🐛 Agent-in-the-Loop response duplication: Resolved a critical bug in the AgentDispatcher that previously led to duplicate Agent-in-the-Loop (AITL) responses and unintended repeated execution of orchestration steps. The dispatcher now accurately subscribes only to control events, preventing these issues.
⚡️ Tracing accuracy for AITL events: Improved the timing of span completion for Agent-in-the-Loop interactions, ensuring tracing data is precisely recorded before final responses are published.

Added

✨ Dedicated control event subscription: Introduced AgentNCSubscriber.for_thread_control_events, a new method that allows agents to precisely subscribe only to control events within a thread, enhancing event handling efficiency and reliability.
📄 Robust AITL response verification tests: Added new comprehensive test scenarios to validate the uniqueness of Agent-in-the-Loop responses and orchestration results, significantly improving the stability and correctness of agent workflows.

[v0.274.2] - 2026-04-10 - Web Platform Tenant Context Enforcement

Changed

🔐 Enforced Tenant Scoping for Agent Management: Operations for creating new agent instances and fetching available agent classes or instances now correctly enforce tenant isolation, ensuring agent data is specific to the active tenant.
📚 Improved Tenant Context for Knowledge Management: Database selectors, document uploads, and detailed document views now properly utilize the tenant context, ensuring knowledge base operations are isolated to the specific tenant.
💡 Applied Tenant Scoping to Model and File Access: API calls for fetching LiteLLM models and resolving file URLs now consistently respect the active tenant, enhancing data security and isolation for these resources.
🚀 Ensured Consistent Versioning Across Components: All core packages, including agent, api, backup, bot, core, pipeline, process, and web, have been updated to version v0.274.2 for release consistency.

[v0.274.1] - 2026-04-10 - Streamlined Data Lake Integrations and Document Handling

Added

✨ Introduced automatic MongoDB Document Store configuration: Rclone to Data Lake pipelines now automatically provision and configure a MongoDB document store based on the data lake container name, streamlining data ingestion and indexing.

Changed

🔄 Enhanced RawLoader file reading robustness: Updated the RawLoader to use a more robust byte-based file reading approach with error replacement, improving handling of various file encodings and malformed characters.

Removed

🗑️ Deprecated placeholder_refdocs_factory: The placeholder_refdocs_factory asset has been removed from SharePoint and local filesystem to Data Lake pipelines, simplifying asset definitions and reducing overhead.

[v0.274.0] - 2026-04-08 - Enhanced Tenant Management and API Routing

Added

✨ Introduced MyTenantController: A new API controller providing endpoints for users to view and manage their tenant memberships, including fetching all associated tenants and setting their active tenant.
📄 New Client-Side Tenant Management Composables: Added a suite of Vue composables (useTenant, useTenantPath, useTenantMemberships, useActiveTenantQuery, useSetActiveTenant, useTenantPolling, useTenantReady) to provide robust and reactive tenant context management across the frontend application.
🖼️ Tenant Selection User Interface: Introduced a dedicated /select-tenant page and a TenantSwitcher component, enabling users to easily view and switch between their available tenants.
🚀 Tenant-Aware Root Landing Page: The main landing page (/) now intelligently redirects users based on their tenant access: auto-selecting for single-tenant users, or presenting a tenant selection page for users with multiple tenants.
🛠️ get_tenant_ids_for_user Utility: Added a new utility method to retrieve all tenant IDs a specific user belongs to, supporting tenant management features.

Changed

🔄 API Base URL for Frontend: The frontend's API base URL has been updated from a hardcoded tenant (/api/v1/active) to a generic /api/v1, aligning with the new flexible tenant-scoped routing.
🐛 Improved Frontend Error Handling for Access Denied: Enhanced the client-side API error handling to automatically redirect users to the tenant selection page upon receiving an "Access denied" (403) response, ensuring a smoother user experience.
🍪 Login Redirect URL Preservation: The authentication middleware now stores the intended redirect URL before login, allowing users to return to their specific page (including tenant context) after successful authentication.
🏛️ Global Auth Provider Endpoints: The Auth Provider API endpoints are now consistently non-tenant-scoped, accessible at /api/v1/auth-providers.

Refactor

🧹 Comprehensive API Routing Overhaul: Redesigned API routing to explicitly distinguish between global and tenant-scoped controllers, with tenant-scoped endpoints automatically prefixed by /{tenant_id}.
🚀 Dynamic OpenAPI Schema Generation: Implemented a custom OpenAPI schema hook to dynamically inject the {tenant_id} path parameter into the OpenAPI specification for all tenant-scoped endpoints, ensuring accurate client SDK generation.
⚡️ Consistent Tenant Context in API Calls: Updated almost all frontend API calls and query keys to explicitly include the tenantId, ensuring proper tenant context and more efficient cache invalidation.
🌐 Tenant-Aware Frontend Page Structure: Major restructuring of the frontend's page directory, moving most service-related pages under a /[tenant]/ dynamic route to natively embed tenant context into URL paths.
⚙️ Modular Controller Architecture: Introduced TenantScopedController as a base class for all tenant-specific API controllers, enforcing consistent tenant-aware routing and authentication practices. Many existing controllers (e.g., Agent, Model, User) now inherit from this new base class.
📖 Updated API Documentation for Routing: The API documentation (CLAUDE.md) has been updated to reflect the new tenant-scoped routing paradigm, clarifying how controllers are mounted and how tenant_id is handled.
🛠️ Dynamic Endpoint Discovery Integration: Updated the endpoint discovery services to leverage the new tenant-scoped routing, ensuring that dynamically generated endpoints (e.g., for agent and process instances) correctly incorporate the tenant context.
📍 Unified Authentication Identity Building: Refactored authentication handlers to use a single build_identity method, which correctly resolves user and tenant context based on the incoming request, improving consistency and maintainability.

[v0.273.6] - 2026-04-07 - Empowering Recovery: New System Restore Workflow

Added

✨ Enabled Full System Restore Workflow: Introduced a comprehensive Dagster-based workflow for performing full system restores, allowing recovery from a specified backup timestamp. This new capability orchestrates service shutdowns, validates backup integrity, manages individual service restorations, and restarts all services upon successful completion.
⚙️ Granular Container Lifecycle Management: Implemented a new ContainerLifecycleManager resource to provide precise control over starting, stopping, and health-checking specific groups of containers required by individual services during the restore process.
🧩 Service Dependency Mapping: Established a centralized mapping (SERVICE_DEPS) that defines the container dependencies for each backup service, ensuring correct sequencing and isolation during parallel backup and restore operations.
🛠️ Modular Service Handler Architecture: Introduced a new BackupHandler interface and a factory mechanism to standardize how backup and restore logic is implemented for each service, improving modularity and extensibility for future service integrations.
✅ Backup Integrity Validation: Added a critical pre-restore validation step that checks for the presence of all expected backup files in S3, preventing restore attempts on incomplete or corrupted backups.
🧪 Comprehensive Restore Test Suite: Included extensive new unit tests covering the container lifecycle management, handler factory, restore definitions, and backup validation to ensure the reliability and correctness of the new restore functionality.

Fixed

🐛 Improved Container Stop Logging: Enhanced the accuracy of container stop logging, so that only successfully stopped containers are counted, providing a more reliable report of system state during backup preparations.

[v0.273.5] - 2026-04-07 - Robust Backup System with S3 Integration and Automated Operations

Added

✨ Introduced a comprehensive backup system: The previously placeholder backup service is now fully functional, providing robust data protection.
🚀 Integrated S3 storage for backups: Backups are now securely stored on S3-compatible object storage, configurable via environment variables.
🗓️ Implemented automated daily backups: A new Dagster schedule runs daily at 1 AM (Zurich time) to perform system-wide backups.
🧹 Added intelligent backup retention policies: Old backups are automatically pruned based on configurable retention days, while ensuring a minimum number of backups are always kept.
🗄️ Enabled service-aware backup for core components: Explicit support for backing up critical services including PostgreSQL, Milvus, Neo4j, ClickHouse, Valkey, and NATS.
⚙️ Streamlined configuration management for backup: Sensitive credentials and backup settings are now securely managed using pydantic-settings and exposed via Docker environment variables.
🔄 Dynamic partitioning for easier restore selection: Dagster now dynamically partitions backups by timestamp, simplifying the selection of a specific backup for restoration.
🌐 Dedicated 'storage' network for backup service: The backup service now connects to a specific storage network in Docker Compose, improving network isolation and access control.

Changed

🛠️ Enhanced backup session robustness: The backup process now automatically purges any pre-existing, incomplete backup artifacts for the current timestamp on S3 before starting a new session.
📄 Improved backup logging clarity: Backup session logs now explicitly include the S3 bucket and prefix for easier monitoring and debugging.

Fixed

🐛 Prevented Traefik service interruption during backup: The backup process now correctly excludes traefik containers from being stopped and restarted, ensuring continuous inbound traffic handling.

Removed

🗑️ Removed placeholder status from backup README: The README for the backup package no longer states it's a skeleton, reflecting its new, fully operational status.

Refactor

🚰 Streamlined Dagster resource injection for backup: Refactored backup-related resources into a unified factory for improved modularity and dependency management.
🧱 Centralized service mapping for backup assets: Defined a clear mapping between backup service names and their corresponding Dagster asset keys, enhancing code organization.
📦 Updated backup package dependencies: Added boto3 for S3 interaction and pydantic-settings for robust configuration.

[v0.273.4] - 2026-04-07 - Improved Authentication and Service Configuration

Added

✨ Keycloak Client Plugin: Introduced a new client plugin dedicated to Keycloak API interactions, facilitating more robust authentication management flows.
💾 LiteLLM Model Persistence: Enabled the STORE_MODEL_IN_DB configuration for LiteLLM services, allowing models to be stored and managed within the database.

Changed

🔐 Enhanced Logout Security: Improved the logout process to explicitly revoke the Keycloak session using the refresh token, ensuring a more complete and secure session termination.
🔄 Forced Login Prompt: Modified the authentication flow to include prompt: 'login', which will now consistently prompt users for their credentials during login.
⚙️ Flexible Keycloak Configuration: Updated Keycloak realm and identity provider template variables to use config_variant_suffix, enhancing configuration flexibility across different deployment stages.
🌐 Open WebUI CORS Delimiter: Adjusted the CORS_ALLOW_ORIGIN configuration in Open WebUI services to use a semicolon (;) as a delimiter for multiple origins, improving compatibility.
📄 Jupyter Lab Boolean Format: Standardized the JUPYTER_ENABLE_LAB setting to use the boolean true format for consistency with Jupyter's configuration expectations.

Removed

🗑️ Development Configuration Cleanup: Cleaned up obsolete blank lines and comments in development-specific docker-compose files, enhancing configuration readability.

[v0.273.3] - 2026-04-07 - OpenWebUI Integration & Agent Provisioning Enhancements

Added

✨ OpenWebUI Agent Workspace Integration: Introduced comprehensive integration with OpenWebUI, allowing dynamic provisioning and management of AI-Hub agent instances as workspace models. This enables a seamless experience for users interacting with AI-Hub agents directly within OpenWebUI.
🔑 Keycloak Role-Based Access for OpenWebUI: Configured Keycloak to include user roles in OpenID Connect tokens and defined specific admin roles for OpenWebUI, enhancing security and access management for the new integration.
⚙️ Automated OpenWebUI Service Account Provisioning: Implemented an automatic creation of a dedicated AI-Hub service account with administrator privileges in OpenWebUI's database, facilitating secure and programmatic model management via API.
🧪 Improved Agent Provisioning with Redis Caching: Enhanced agent discovery and synchronization to external platforms (Langfuse, OpenWebUI) by implementing a Redis-backed, hash-based caching mechanism, reducing redundant sync operations and improving efficiency.
🔐 New JWT Library for OpenWebUI Security: Added pyjwt as a new core dependency, providing essential cryptographic capabilities for secure token-based communication with OpenWebUI APIs.

Changed

🔄 Standardized Agent and Process Health Check Ports: Unified the default health check port for all agent and process runners to 8090 (previously 8080), preventing potential port conflicts and ensuring consistent monitoring.
🚀 Refactored External Provisioning Orchestration: Streamlined the agent instance synchronization logic within the API service, enabling a more robust and extensible system for provisioning agents to platforms like Langfuse and OpenWebUI.

[v0.273.2] - 2026-04-07 - Establishing the Backup Service Framework

Added

✨ New Automated Backup Service: Introduced a dedicated backup service built on Dagster, designed to orchestrate and manage system backups. This service runs as an independent set of Docker containers within the main project.
🦾 Container Orchestration Capabilities: Added core functionality to discover, stop, and restart Docker Compose containers using direct Docker daemon interaction, forming the basis for coordinated backup and restore operations.
⚡️ Resilient Failure Recovery: Implemented a critical safety mechanism that automatically restarts all managed containers if any part of the backup process fails, preventing prolonged service downtime.
📦 Comprehensive Infrastructure for Backup Service: Established all necessary Docker Compose configurations, Dagster definitions, and tooling for deploying and managing the new backup service across various environments (dev, build, local, latest, nightly).
📄 Docker Interaction Utilities: Provided a DockerManager for low-level Docker operations, enabling precise control over container lifecycle, file transfers, and command execution within the backup service.
🚦 Dedicated Backup UI: The backup Dagster instance now includes a dedicated webserver, accessible at http://localhost:3004 in dev and local environments, for monitoring and manual triggering of backup processes.

Changed

🔄 Main Dagster Logging Configuration: Updated the main Dagster instances to log swiss_ai_hub messages at INFO level, providing more consistent operational insights.
🧹 Streamlined Backup Service Configuration: Moved the backup service's dagster.yaml and workspace.yaml configurations into dynamically generated Docker Compose volume mounts, simplifying deployment and ensuring consistency.

Removed

🗑️ Deprecated Backup Development Workflow: Removed the standalone dagster dev command from the backup package's Makefile, as development is now integrated with the Docker Compose setup.

[v0.273.1] - 2026-04-01 - New Automated Backup Service & Improved Build Processes

Added

✨ Introduced Backup Service: A new Dagster-based service has been added to provide automated database backup orchestration. This initial release establishes the foundational structure for future backup capabilities.
🔒 Configured Backup Service Authentication: Integrated Keycloak OAuth2 Proxy for the new Backup Service, ensuring secure access and management through a dedicated endpoint (e.g., backup.<DOMAIN>).
📦 Added Docker Image Build Workflow for Backup: A new GitHub Actions workflow has been established to automate the building and releasing of the backup service Docker image.
⚙️ Integrated Backup Service into Build System: The new backup package is now fully integrated into the project's Makefile for testing, linting, formatting, PR readiness checks, and consistent version bumping.
🚀 Deployed Backup Service via Docker Compose: The backup service, comprising Dagster code server, webserver, and daemon, along with its OAuth2 proxy, is now deployable across all environments via updated Docker Compose configurations.
📄 Initial Backup Service Documentation: A README.md file has been added to the backup package, outlining its purpose as a placeholder for backup and restore orchestration.

Refactor

🧹 Standardized Docker Build Contexts: Refined the Dockerfile build contexts for numerous existing services (e.g., api, web, agents, pipelines, bot) in the Docker Compose configurations to ensure consistent and correct local image building across the monorepo structure.

[v0.273.0] - 2026-04-01 - Multi-Tenant API Routing & Active Tenant Experience

Added

🦾 PR Demo Video Skill: Introduced a new skill that automates the generation of pull request demo video descriptions and updates test plan checklist items based on screen recordings, streamlining the PR review process.
✨ Active Tenant Persistence: The UserEntity now includes an active_tenant_id field, allowing the system to remember a user's last selected tenant for an improved multi-tenant user experience.
📄 Tenant Path Parameter ADR: A new Architecture Decision Record (2026_03_30_tenant_path_parameter.md) was added, detailing the strategic shift to tenant identification via a required URL path parameter.

Changed

🔄 Core API Tenant Routing: All primary API endpoints now require a {tenant_id} path parameter (e.g., /api/v1/{tenant_id}/...) for explicit tenant context. The x-tenant-id HTTP header is no longer supported, and there is no implicit fallback to a default tenant.
🚀 "Active" Tenant Slug: Introduced a special "active" slug (e.g., /api/v1/active/...) that resolves to the authenticated user's persisted active tenant, simplifying frontend interactions.
🔐 Authentication Handlers: Updated AuthHandler and its implementations (Keycloak, DangerousDevelopmentOnly, Token) to resolve tenant context from the URL path parameter and use the active_tenant_id for contexts without a request (e.g., WebSocket connections).
🤖 Bot Framework Endpoints: Azure Bot Service and other bot connection configurations were updated to utilize the new tenant-scoped API paths, primarily through the /api/v1/active/... slug.
🌐 Frontend API Client: The main frontend API client's baseURL was updated to use /api/v1/active, ensuring all frontend-initiated requests correctly leverage the active tenant context.
⚙️ Deployment Configurations: Numerous environment variables across docker-compose files (dev, build, latest, local, nightly variants) were updated to point to the new /api/v1/active/... API paths, ensuring consistent deployments.
📚 Documentation & Skill Guides: Extensive updates were made across all relevant documentation (ADRs, platform docs, skill definitions, CLAUDE.md, README.md) to reflect the new API routing, tenant identification, and active tenant behavior.
🩺 Health Endpoints: Health check endpoints are now specifically mounted outside the tenant scope at /api/v1/health/, allowing for unauthenticated status checks.

Fixed

🐛 Tenant Data Consistency on Deletion: Enhanced the delete_tenant_by_id and remove_user_from_tenant operations to automatically clear the active_tenant_id for affected users, preventing stale tenant references when tenants or user-tenant associations are removed.

Refactor

🧹 API Routing Implementation: The core ApiRunner logic was significantly refactored to implement the new tenant-scoped routing using Starlette Mount for path parameter capture, and to separate health endpoints.
🧪 Test Suite Alignment: Various API and authentication test cases, including mock request creation and tenant resolution scenarios, were updated to align with the new tenant path parameter and active tenant resolution logic.
⚡️ OpenTelemetry Tracing: Improved the OpenTelemetry span attributes in Controller to accurately record the resolved tenant ID for business context, filtering out the raw {tenant_id} path parameter.

[v0.272.2] - 2026-03-31 - Enhanced Configuration Authorization for Secure Operations

Added

✨ Introduced Configuration Authorization Service: A new service (ConfigAuthorizationService) has been implemented to automatically validate user permissions for resources referenced within agent and process configurations (e.g., knowledge databases, other agents) during their creation and update.
📄 Authorization Violation Model: A new data model (ConfigAuthorizationViolation) was added to standardize the reporting of configuration authorization issues, providing clear details on unauthorized resources.
🌐 Localized Authorization Messages: New translation keys for configuration authorization messages have been added across all supported languages (German, English, French, Italian), ensuring a consistent user experience.

Changed

🔑 Mandatory Configuration Authorization: Agent and Process instance creation and update operations now include a mandatory step to enforce authorization checks on submitted configurations, preventing users from referencing resources they do not have access to.
🦾 Agent Selector Authorization: The Agent Selector form element now performs real-time authorization validation, ensuring users can only select and configure agents they are permitted to access.
📚 Knowledge Database Selector Authorization: The Knowledge Database Selector form element now performs authorization validation, ensuring users can only select knowledge databases they are permitted to access.
🔄 Model Configuration Update: The default Claude model has been updated to opus[1m], and explicit support for this new model variant has been added to the list of available models.

Refactor

🧹 Form Module Exports: Key form components, including ConfigAuthorizationViolation, FormkitElement, Group, ModelSelect, and Repeater, are now explicitly exported within the core form module for improved internal consistency and module organization.

[v0.272.1] - 2026-03-30 - Robust Agent Context: Valkey AOF and True Thread Persistence

Fixed

⚡️ Enhanced Valkey Data Durability: Addressed the previously identified "Valkey persistence gap" by enabling Append-Only File (AOF) persistence across all Valkey instances. This crucial update reduces potential data loss during crashes from 30 seconds to approximately 1 second, significantly improving the reliability of agent runtime state.

Changed

💾 Introduced Truly Persistent ThreadContext: Modified the ThreadContext to be genuinely persistent by removing its 30-day Time-to-Live (TTL). This change ensures that conversation history, user preferences, and other long-lived agent-set data remain durable across multiple runs without automatic expiry.
⚙️ Refined RunContext Lifecycle: Clarified the behavior of RunContext, which is now explicitly cleaned up upon StopEvent completion. Its 30-day TTL now functions as a safety net for orphaned runs in case of unexpected crashes, ensuring data integrity while simplifying cleanup.
📄 Updated Agent Context Documentation: Improved documentation across CLAUDE.md, solution strategy, and SDK guides to accurately reflect the new persistence models and durability guarantees for RunContext and ThreadContext in Valkey.

Refactor

🧹 Improved IDE Configuration: Streamlined the Node.js interpreter setup in Web.xml to use project-level settings, enhancing consistency for developers.
🧹 Internal Code Cleanups: Performed minor code reordering for imports and test configurations to improve maintainability and style.

[v0.272.0] - 2026-03-25 - Empowering Agents with External Tooling: Introducing the MCP React Agent

Added

🦾 New MCP React Agent: A sophisticated AI agent capable of dynamically discovering and interacting with tools and resources exposed by external Message Control Protocol (MCP) servers, leveraging ReAct reasoning.
🚀 MCP Integration Framework: Core components and utilities (McpClientFactory, McpClientConfig, mcp_resource_schemas, mcp_tool_schemas) for seamless and robust communication with MCP-compliant services.
📄 Architectural Decision Record (ADR): A detailed document outlining the strategy for agents to consume MCP resources, utilizing system prompts for static data and a meta-tool for templated data.
🌐 Internationalization for MCP Agent: Added comprehensive localization support for the MCP React Agent's metadata, configuration parameters, and workflow steps in German, English, French, and Italian.
✨ Tool Call ID to ToolEvent: Introduced a tool_call_id field within ToolEvent to establish a clear link between a tool invocation and its corresponding result message.
🐳 Docker Deployment for MCP Agent: Provided a Dockerfile to facilitate the easy containerization and deployment of the new MCP React Agent.
🧪 Playground Workflow for MCP React Agent: A minimal example workflow, complete with a dummy MCP server, to demonstrate and test the capabilities of the MCP React Agent.
📦 fastmcp Dependency: Integrated fastmcp as a crucial dependency to enable direct client interaction with MCP servers.

Changed

🔄 Enhanced Message Tool Call Handling: Improved the Message class to accurately normalize and preserve tool_calls data during conversion to and from llama_index's ChatMessage format, ensuring data integrity across the LLM pipeline.
🔗 OpenAI API Compatibility for Message: Added methods to_openai_dict and from_openai_response to the Message class, enabling better interoperability with OpenAI API messages, especially for complex tool interactions.
🛡️ Robust Content Extraction in Message: Made the content property of the Message class more resilient by explicitly handling cases where message contents might be empty.

Refactor

🧹 Organized MCP Core Components: Restructured the codebase by creating dedicated packages/agent/mcp/ and packages/core/mcp/ directories, centralizing all MCP-related infrastructure and agent-specific logic.
⚙️ Standardized Line Endings: Ensured consistent LF line endings for .sh files across the repository via .gitattributes for improved cross-platform compatibility.

Changed

⚡️ Improved Document Intelligence Loader: Enhanced file processing within the Document Intelligence Loader by switching to direct byte content reading, leading to more robust and compatible document analysis.

[v0.271.5] - 2026-03-24 - Documentation Base Path Update and Link Alignment

Changed

📄 Updated Documentation Base Paths: The base path for all project documentation has been consistently updated from /swiss-ai-hub/ to /aihub-core/ across the VitePress configuration, README.md, CONTRIBUTING.md, CLAUDE.md, and the install.sh script. This ensures all documentation links and internal paths correctly point to the new aihub-core URL structure.

[v0.271.4] - 2026-03-23 - Enhanced AI Workflow Automation and API Resilience

Added

🦾 Introduced a new splice-issue skill for the AI assistant, enabling it to intelligently decompose large GitHub issues into smaller, independently mergeable sub-issues with appropriate dependency relationships, metadata inheritance, and project board priority propagation.

Changed

📄 Improved license report generation by making the generate-license.sh script smart enough to preserve the existing "Generated on" date if the core content of the license report has not changed, reducing noise in version control.
⚙️ Increased LiteLLM rate limit error retries from 3 to 5 across all proxy configurations, enhancing the system's resilience and reliability against temporary API rate limits from model providers.

[v0.271.3] - 2026-03-23 - Improved Agent Docker Image Documentation

Changed

📄 Enhanced Agent Docker Images: The README.md file from the packages/agent module is now included in the Docker builds for all agent applications, ensuring better in-image documentation and context.

[v0.271.2] - 2026-03-23 - Enhanced Test Suite Reliability

Changed

⚡️ Improved Test Suite Stability: Marked several integration and memory-related tests as flaky, ensuring automatic retries on transient failures to enhance overall CI/CD reliability and reduce false-positive test failures.

Fixed

🐛 Development Configuration File: Corrected the JSON string formatting for OTEL_CLOUD_HEADERS within the .env.dev file to ensure proper parsing.

Changed

📄 Changelog Accuracy: Updated the changelog entry for v0.271.0 to fully detail the significant "Grand Renaming and Open-Source Restructure" that took place in that release.
⚙️ Frontend Linting Process: Modified the frontend linting GitHub Action to include an additional nuxi prepare .app step, ensuring comprehensive preparation for linting in the updated project structure.
🤖 ChatBot Test Setup: Enhanced ChatBot tests to provision a default tenant and assign administrative roles to the test user, improving the realism and robustness of multi-tenancy and authorization test environments.

Refactor

🧹 CI/CD Artifact Naming: Standardized the naming convention for build artifacts (pytest, coverage reports) across GitHub Actions workflows by sanitizing working directory paths, increasing robustness and consistency in artifact management.
🔄 CI/CD Workflow Paths: Updated internal GitHub Actions workflow references from the older swiss-ai-hub path to aihub-core to align with the new repository and package naming conventions.
🏷️ SonarCloud Project Identifiers: Aligned SonarCloud project keys and names across all packages (agent, api, bot, core, pipeline, process, web) to reflect the new aihub-core branding and simplified naming.

[v0.271.0] - 2026-03-19 - Grand Renaming and Open-Source Restructure

Changed

🏗️ Repository Layout: Reorganized all source packages under packages/ directory — aihub_lib → packages/core, aihub_agent → packages/agent, aihub_api → packages/api, aihub_bot → packages/bot, aihub_pipeline → packages/pipeline, aihub_process → packages/process, swiss_ai_hub_web → packages/web. Non-code directories moved to conventional locations: aihub_doc → docs, aihub_action → .github/actions, deployment → infra/deployment.
📦 Python Namespace Packages: All packages renamed to use swiss_ai_hub namespace via uv_build — import paths are now swiss_ai_hub.core, swiss_ai_hub.agent, etc. PyPI names follow as swiss-ai-hub-core, swiss-ai-hub-agent, etc.
🐍 snake_case Everywhere: Renamed all CamelCase Python source files and directories to snake_case (e.g., ExpertAskingAgent/ExpertAskingAgent.py → expert_asking_agent/expert_asking_agent.py), resolving import ambiguity that previously prevented proper __init__.py usage.
📤 Public Interface via __init__.py: Added lazy __init__.py exports (using TYPE_CHECKING + __getattr__) to all top-level package folders, establishing clear public interfaces. Cross-package imports now go through __init__.py; intra-package imports use full module paths.
🔄 Import Rewrite: Updated all imports across the entire codebase to follow the new namespace structure and import conventions.
📝 License Change: Moved from LicenseRef-Proprietary to Apache-2.0 across all packages.
📚 Documentation Updates: Updated all references from old naming conventions (aihub_doc, aihub-core, etc.) to new names across docs, READMEs, CLAUDE.md files, whitepapers, and configuration files.
⚙️ Docker & CI Updates: Updated all Docker Compose files, Dockerfiles, GitHub Actions workflows, and run configurations to reference the new directory structure and package names.

Added

📄 Architecture Decision Record: Added ADR documenting the rename and restructure rationale (docs/arc42/decisions/2026_03_13_rename_and_restructure_for_open_source_release.md).
🏗️ docker-compose.build.yml: Added comprehensive build-stage Docker Compose for multi-stack orchestration.

Removed

🗑️ Old Directory Structure: Removed all aihub_* top-level directories, replaced by packages/ layout.
🗑️ Deprecated Test Suites: Removed outdated test suites for AgentConfig and Form classes.
🗑️ Outdated READMEs: Removed stale README files that were superseded by the restructure.

[v0.270.4] - 2026-03-19 - Enhanced Configuration and Dev Environment Streamlining

Changed

✨ Updated MinerU2.5-2509-1.2B Model Identifier: Corrected the internal model name from openai/mineru to openai/MinerU2.5-2509-1.2B across all LiteLLM configurations, ensuring accurate model referencing.
⚙️ Simplified Development CORS Configuration: Adjusted the CORS_ALLOW_ORIGIN setting for development environments to http://localhost:8080, streamlining local frontend development and testing.
🌐 Improved Development Playwright Integration: Updated the PLAYWRIGHT_WS_URL for development environments to ws://localhost:3036, making local Playwright service integration more straightforward.

[v0.270.3] - 2026-03-11 - Repository Cleanup: Streamlining Project Configuration

Removed

🗑️ IDE Configuration Files: Deleted all .idea/ directories and their associated configuration files (e.g., misc.xml, modules.xml, inspectionProfiles/) from the main project and various sub-projects (aihub_agent, aihub_api, aihub_bot, aihub_lib, aihub_pipeline, aihub_process) to clean up the repository and remove environment-specific configurations.

Changed

🧹 Updated Git Ignore Rules: Modified the root .gitignore file to explicitly exclude .idea/misc.xml and all aihub_*/.idea/ directories, preventing IDE-specific files from being committed in the future and maintaining a cleaner codebase.

[v0.270.2] - 2026-03-09 - Deployment Script Readability Enhancements

Refactor

🧹 Enhanced Deployment Script Clarity: Renamed internal variables for destination paths (output_dir to dst_rel) within the deployment generation scripts to improve code readability and maintainability.

[v0.270.1] - 2026-03-05 - Enhanced Authentication Experience and Deployment Flexibility

Added

✨ Custom Keycloak Login Theme: Introduced a new aihub custom login theme for Keycloak, providing a branded and dark-mode compatible authentication experience.
🔑 Keycloak Account Audience Mapper: Added an OIDC audience mapper for the account client within Keycloak realms, ensuring proper audience claims in tokens.

Changed

🚀 Flexible Keycloak Entrypoint: Updated the Keycloak entrypoint script and its Docker Compose integration to support dynamic arguments for kc.sh, enhancing deployment flexibility across all stages.
⚙️ Standardized Keycloak Realm Imports: All Keycloak realm configurations are now consistently processed as templates by the entrypoint script before import, streamlining environment variable substitution.
🌐 Updated Development Environment URLs: Adjusted Keycloak client rootUrl, redirectUris, and post.logout.redirect.uris for the development environment to http://localhost:3333, aligning with current frontend settings.
🔐 Refined OpenWebUI OIDC Configuration: Modified the OPENID_PROVIDER_URL for OpenWebUI in development environments to http://localhost:8180 for improved local setup consistency.
🛠️ Enhanced Deployment Script for Static Assets: The generate_compose.py script now supports copying static directories and files, enabling the seamless inclusion of custom themes and other assets in deployments.
📝 Adjusted Logging Configuration: Fine-tuned the list of modules for which logging is disabled in aihub_lib for improved log output.

Fixed

🐛 License Generation Locale Compatibility: Added LC_ALL=C export to the generate-license.sh script to ensure consistent behavior across different locales.

[v0.270.0] - 2026-03-04 - Keycloak Takes the Helm: Major Authentication Overhaul with Dynamic Identity & Enhanced Security

Added

🔐 Keycloak Identity and Access Management (IAM) System: Fully integrated Keycloak as the central identity broker, enabling OIDC/OAuth2 authentication for all platform services. This provides multi-provider federation, improved local development experience, and enhanced control over identity.
✨ Dynamic Identity Provider Discovery: Introduced an API endpoint (/auth-providers) and a new frontend component to dynamically fetch and display available identity providers from Keycloak. This allows for flexible, IdP-specific login buttons with custom icons on the authentication page, eliminating static configuration.
📄 New Architecture Decision Records (ADRs): Documented key architectural decisions regarding Keycloak integration, seamless OpenWebUI SSO in iframes, parent application ownership of authentication state, and multi-tenant assignment via Keycloak groups.
🧪 Playwright E2E Testing Framework: Implemented a new End-to-End (E2E) testing framework using Playwright for the web UI, including an automated authentication setup to ensure reliable testing of user flows.
⚙️ Centralized Tenant Configuration: Introduced a new TenantSettings class to centralize configuration for multi-tenant aspects, such as default tenant name, description, access rules, and initial user signup roles.
➕ Dedicated Keycloak API Service Account: Created a least-privilege Keycloak service account (aihub-api-service) for the API backend to securely query identity provider information.

Changed

🔄 Unified OIDC Client Configuration: Updated the platform's OIDC client to directly integrate with Keycloak, ensuring consistent authentication against the new identity broker.
🛠️ Service-Wide Authentication Rework: Reworked the authentication configuration across all Docker Compose services (Admin UI, OpenWebUI, Dagster, SeaweedFS, Attu, Langfuse, API) to leverage Keycloak as the OIDC provider, ensuring a consistent and centralized authentication flow.
🚀 Improved OpenWebUI SSO Experience: Optimized the integration of OpenWebUI by directly initiating its OAuth login flow within the iframe, providing a seamless Single Sign-On experience for users.
🌎 Dynamic Login Page UI: The frontend login page now dynamically renders authentication buttons based on the discovered identity providers, enhancing flexibility and user choice.
📝 Comprehensive Documentation Updates: Updated platform documentation, including the authentication setup, prerequisites, multi-tenancy guides, and quick start instructions, to reflect the new Keycloak-based identity management system and associated roles.
📦 PostgreSQL Database Expansion: Configured the PostgreSQL service to manage a new dedicated database for Keycloak, ensuring proper data isolation and persistence for the IAM system.
🏗️ Automated Compose File Generation: Enhanced the generate-compose Makefile target to automatically include Keycloak-specific configurations and ensure proper YAML formatting.

Security

🛡️ Elevated Admin Tool Access Role: Restricted access to critical infrastructure administration tools (Dagster, SeaweedFS, Attu) to users with the new AIHubSysAdmin realm role, enhancing security for these sensitive services.
🔒 Hardened Keycloak Admin Console Access: Documented and implemented IP allowlisting capabilities via Traefik middleware to restrict access to the Keycloak admin console and metrics endpoint in production deployments.
🔐 Secure OpenWebUI CORS Policy: Hardened the Cross-Origin Resource Sharing (CORS) policy for OpenWebUI, explicitly allowing access only from trusted domains to prevent unauthorized access.
👮 Mandatory AIHubAccess Role Enforcement: Configured Keycloak authentication flows to deny login to any user who does not possess the AIHubAccess realm role, acting as a crucial access gate.

Fixed

🐛 Graceful Iframe Logout Handling: Resolved an issue where logging out from embedded services (e.g., OpenWebUI) could lead to a blank iframe. The parent application now detects this and redirects the user to the home page while preserving the main session.
🩹 S3 Anonymous File Download Error Handling: Corrected the error type from IOError to OSError in the S3 anonymous file access service, improving the robustness and accuracy of error handling during file download operations.

Removed

🗑️ Direct Azure AD Configuration & Handlers: Eliminated all environment variables and code paths related to direct Azure AD integration, including the OAuth2AuthHandler and its settings, in favor of the new Keycloak-centric architecture.
🗑️ Obsolete Authentication Strategy Resolver: Eliminated an outdated utility for resolving authentication strategies, simplifying the authentication codebase.

[v0.269.4] - 2026-03-03 - Streamlined Setup and Enhanced Observability

Added

🚀 One-Command Installer: Introduced a new install.sh script to simplify platform setup and upgrades, featuring auto-detection for GPU hardware and robust .env secret generation.
🖼️ Platform Tour Demos: Added new demo videos and images to the README.md and documentation, visually showcasing key platform features like document ingestion, agent interaction, and cost control.
✨ Standardized Startup Banner: All core services now display a consistent startup banner with version information, improving visibility and branding upon service launch.

Changed

📄 Revamped Main Documentation: The README.md has been completely rewritten to provide a higher-level overview, a more intuitive quick start guide, and a comprehensive platform tour, making it easier for new users to understand and engage with the Swiss AI-Hub.
⚙️ Unified Versioning Configuration: Renamed the AIHUB_API_VERSION environment variable to AIHUB_VERSION across all .env files, Dockerfiles, and internal settings for greater consistency.
📦 Centralized Version Management: The platform's version is now automatically derived from package metadata, ensuring consistency across all components and deployment artifacts.
🇩🇪 Improved German Documentation: Enhanced clarity and detail in German documentation, particularly for sections on low-level traces (including AITL delegation details), quick start, and update/maintenance procedures.

Refactor

🧹 Consolidated Version Environment Variables: Standardized the version environment variable name (e.g., AIHUB_AGENT_VERSION, AIHUB_PIPELINE_VERSION merged into AIHUB_VERSION) across all Python services and Docker images for simplified configuration.

[v0.269.3] - 2026-03-03 - Expanded Document Parsing and Flexible File Handling

Added

✨ New RawLoader for Plaintext Files: Introduced a dedicated loader for various plaintext formats (e.g., txt, md, csv, json, xml, yml, yaml, log, ini, cfg, toml) to ensure accurate content parsing without conversion.
⚙️ Configurable Passthrough Extensions: Added a new setting, ParsingSettings.PASSTHROUGH_EXTENSIONS, allowing configuration of file types (like zip or wav) that bypass markdown conversion and return empty content, enabling agents to access the raw files directly from S3.
📦 Direct S3 File Download: Implemented S3AnonymousFileAccessService.download_file to enable raw byte downloads from S3, facilitating in-memory processing for various file types beyond standard document parsing.

Changed

🔄 Updated Document Parsing Service: The ParsingService now intelligently routes documents to the new RawLoader for plaintext files and incorporates the configurable passthrough mechanism for agent-centric file types, significantly improving versatility and robustness.
📄 Dynamic Extension Support in Parser: The DocumentParserResource now dynamically includes all supported extensions from RawLoader, simplifying configuration and ensuring consistent parsing capabilities.
🚨 Improved Unsupported File Type Messaging: Enhanced the error message for unsupported file types in the parsing service to clearly list all currently recognized extensions, including those handled by the new RawLoader.

Refactor

🧹 Minor Import Clean-up: Removed an unused import in the TraceStore module for improved code hygiene.

Fixed

📦 Improved Release Archive Structure: Corrected the packaging process for .tar.gz release artifacts to ensure they now extract into a proper, versioned top-level directory, enhancing consistency and user experience upon extraction.

[v0.269.1] - 2026-02-27 - Enhanced Distributed Tracing and Agent-in-the-Loop Observability

Added

🚀 Introduced TraceStore: A new Redis-backed store designed to persist tracing metadata, which is crucial for enabling distributed tracing across agent runs and Agent-in-the-Loop (AITL) delegations.
✨ Added OpenInference Trace Context Propagation: Implemented a new context management system (openinference_trace_context) that propagates OpenInference trace state. This ensures that CHAIN span kinds are automatically applied to traced functions, maintaining consistent observability across workflows.
🦾 Implemented AITL Wrapper Spans: New tracing logic creates dedicated AGENT wrapper spans for Agent-in-the-Loop (AITL) delegations. These spans bridge the caller's trace to the delegated agent's steps, enabling clear nested visibility in observability tools like Langfuse.

Changed

🔄 Refactored Agent Tracing Architecture: The AgentRunTracer was significantly refactored to utilize the new TraceStore for all run metadata, including user input, user ID, and LLM output. This change replaces previous in-memory caches and enables cross-service trace continuity.
📝 Updated Tracing Documentation: The platform's documentation on low-level traces has been thoroughly updated to reflect the new distributed tracing architecture, AITL delegation, and enhanced Langfuse integration details.
⚙️ Improved OpenTelemetry Collector Configuration: Modified OpenTelemetry collector configurations to include a filter/noise processor in the Langfuse trace pipeline. This helps reduce unwanted telemetry noise and improves the quality of ingested traces.
⚡️ Enhanced Docker Compose Dependencies: Adjusted docker-compose configurations to improve service startup reliability by ensuring that the langfuse-server service now explicitly waits for langfuse-web to be in a healthy state.

Fixed

🐛 Corrected Retriever Namespace Filtering: Addressed an issue in filter_retrievers_by_namespace where retriever namespaces were being updated incorrectly. The fix ensures proper filtering based on the vector_store's index_namespaces.
🛡️ Improved Langfuse Model Provisioning Robustness: Enhanced the LangfuseProvisioner to gracefully handle 400 errors containing "already exists" messages during model definition creation, treating them as conflicts similar to 409 status codes to prevent failures.

Refactor

🧹 Streamlined AgentDispatcher Cleanup: Simplified the run completion cleanup logic within the AgentDispatcher. The explicit clear_run call for tracing metadata is now removed, as cleanup is automatically handled by the TraceStore.
🧪 Overhauled Agent Tracing Test Suite: The entire test suite for AgentRunTracer has been updated to align with the new Redis-backed TraceStore and the distributed tracing functionalities, removing outdated cache tests and adding comprehensive coverage for AITL scenarios.
⚙️ Refined Python Auto-Formatting Hook: Adjusted the .claude/hooks/auto-format-python.sh script to ignore the F401 (unused import) ruff check during the --fix step, providing more flexibility for specific code patterns without causing formatting conflicts.

[v0.269.0] - 2026-02-26 - Revamped AI Model Strategy: Embracing Swiss Sovereignty with Dual-Mode Inference & Standardized Models

Added

✨ Introduced Swiss LLM Cloud Integration: Full support for Swiss LLM Cloud across various AI tasks (text generation, embedding, reranking, whisper, OCR) to ensure data sovereignty and provide a diverse range of models for non-GPU deployments.
🦾 Integrated Local vLLM for GPU Deployments: Replaced llama.cpp with vLLM for high-performance local GPU inference, supporting text generation, embeddings, and reranking on NVIDIA RTX 6000 Pro with explicit VRAM budgets.
📄 New Architectural Decision Record (ADR): Added a comprehensive ADR outlining the rationale and details of the Swiss LLM Cloud and local vLLM dual-mode inference strategy.
💡 New AI Toolkit Settings: Added .idea/ai_toolkit.xml for improved AI-assisted development experience in IDEs.
⬆️ New Environment Variables for Swiss LLM Cloud: Added specific environment variables for different Swiss LLM Cloud API endpoints (embedding, reranking, whisper, OCR) for fine-grained configuration.

Changed

🔄 Centralized AI Model Strategy: Migrated the platform's core AI model inference from Azure OpenAI and Cohere to a dual-mode strategy utilizing Swiss LLM Cloud for non-GPU deployments and local vLLM for GPU-enabled setups.
⚙️ Standardized Model Naming Convention: Replaced abstract model tier names (e.g., text-generation/nano, embedding/large) with canonical, descriptive model names (e.g., text-generation/gpt-oss-120b, embedding/bge-m3) across LiteLLM configurations, agent templates, and pipeline definitions for clarity and consistency.
📝 Updated Documentation for Model Strategy: Extensively updated architecture, quick start, deployment guide, and security documentation to reflect the new AI model inference strategy, including details on Swiss LLM Cloud and vLLM integration.
🚀 Enhanced Document Parsing with MinerU: Transitioned the primary document parsing service from Docling to MinerU, updating related configurations, resource definitions, and documentation in Dagster pipelines and API routes.
⚡️ Improved Mem0 Compatibility: Integrated patches for Mem0's OpenAI LLM and embedding components to ensure seamless compatibility with Swiss LLM Cloud's API behavior, particularly concerning dimensions and response_format.
🔢 Adjusted Milvus Embedding Dimensions: Updated the default Milvus embedding dimension from 3072 to 1024, aligning with the BGE-M3 embedding model used in the new model strategy.
🗣️ Refined Expert Escalation Prompt: Modified the expert_answer_sufficient prompt template to focus on the latest expert answer in the conversation history, improving the accuracy of agent-driven expert interaction assessments.
🛠️ Updated GitHub Actions for CI: Modified the analyze-test-pr.yml workflow to remove llama.cpp services, inject new Swiss LLM Cloud secrets, and no longer skip azure-marked tests, streamlining the CI pipeline.
🧹 Updated IDE Test Configurations: Removed the -k "not azure" flag from PyCharm/IntelliJ run configurations for API, Agent, Bot, Lib, and Process tests, enabling all tests to run without Azure-specific exclusions.

Fixed

🐛 Ensured Idempotent Logging Configuration: Implemented a fix in aihub_lib's logging setup and OpenTelemetry integration to prevent duplicate log handlers from being added when enable_logging is called multiple times.

Removed

🗑️ Deprecated llama.cpp Inference Services: Eliminated llama.cpp containers for chat, embeddings, and reranking from Docker Compose files, in favor of vLLM for GPU deployments or Swiss LLM Cloud for non-GPU.
🗑️ Removed Azure OpenAI and Cohere Integrations: Azure OpenAI base URLs and keys, as well as Cohere API base and keys, were removed from environment configurations and LiteLLM settings, consolidating cloud inference to Swiss LLM Cloud.
🚫 Dropped Image Generation and Text-to-Speech (TTS) Capabilities: Removed DALL-E 3 and Kokoro/TTS model configurations due to the lack of Swiss-sovereign alternatives in the new model strategy.
🧹 Eliminated Azure-Specific Test Markers: The @pytest.mark.azure marker and related exclusions were removed from Python test files and pyproject.toml configurations, simplifying test management.
❌ Removed AI Code Review GitHub Action: The review-pr.yml workflow and its associated action definition were removed from the repository.

Refactor

🧹 Streamlined Agent Configuration: Removed the redundant agent_class field from AgentConfig and its subclasses across various agent definitions in aihub_agent for cleaner configuration.
🔄 Standardized License Generation Output: Modified the generate-license.sh script to sort Python packages and Docker images alphabetically in the output, ensuring deterministic and consistent license reports.
⚙️ Reordered pr-ready Makefile Targets: Adjusted the execution order of format-md and format-yaml in the pr-ready Makefile target for logical consistency.

[v0.268.0] - 2026-02-26 - Introducing Secure Agent File Uploads and Streamlined Integrations

Added

✨ New Secure File Upload API for Agents: Introduced dedicated API endpoints to securely initiate and validate file uploads to agent instances, providing presigned URLs for direct bucket interaction.
🚀 Dedicated Agent File Storage: Implemented a new shared S3 bucket (agent-files) for all agent-related user uploads. This bucket is automatically created and configured with a 7-day lifecycle policy to expire files, ensuring efficient storage management.
🦾 File ID Reference System: Enhanced the UserUploadedFile NATS event model to include a unique file_id, enabling agents to reference uploaded files by identifier instead of directly embedding large file data.
🧪 Comprehensive Test Coverage: Added extensive unit and integration tests to ensure the reliability and security of the new file upload service and API endpoints.

Changed

🔄 OpenWebUI File Upload Workflow: The OpenWebUI integration pipeline has been updated to utilize the new secure agent file upload API. Files are now fetched from OpenWebUI's S3, uploaded to the agent's dedicated bucket via presigned URLs, and validated, streamlining the data flow.
🔒 Granular File Access Control: Agent instance permissions are now tied to file upload capabilities, ensuring that only authorized users can upload files to specific agent instances.

Removed

🗑️ Deprecated file_data in NATS Event: The file_data field (base64-encoded content) has been removed from the UserUploadedFile NATS event, shifting to a more scalable file reference mechanism.
🧹 Internal S3 Storage Adapter from OpenWebUI Pipeline: The internal S3StorageAdapter and FileStorageAdapter protocol have been removed from the OpenWebUI integration, as file interaction is now handled by the new AI-Hub API.

Security

🔑 Path Traversal Prevention: Introduced stringent validation for filenames and path segments in file upload requests and the UserUploadedFile model to actively prevent path traversal attacks.
🛡️ Presigned URL Security Model: Leveraged presigned URLs for file uploads, significantly enhancing security by providing temporary, single-use access to S3 resources, eliminating the need for persistent credentials on the client side.

[v0.267.4] - 2026-02-26 - Pipeline Robustness: Enhanced Partition Key Encoding

Added

✨ New Partition Key Encoding Utilities: Introduced encode_partition_key and decode_partition_key functions to safely URL-encode and decode file paths, ensuring compatibility when used as Dagster partition keys.
🚀 encode_partition_keys Configuration: Added a new encode_partition_keys parameter to observable_data_lake_factory, observable_local_file_system_factory, and observable_rclone_factory, allowing pipelines to opt-in to URL-encoded partition keys.
⚙️ IO Manager Encoding Support: LocalFileSystemIOManager, RcloneIOManager, and S3DataLakeIOManager now include an encode_partition_keys configuration, enabling automatic decoding of partition keys for accurate file retrieval.
🛡️ Robust Data Versioning: Data versioning ops for data lake, local file system, and rclone sources (data_version_by_partition_for_data_lake_files_no_op, data_version_by_partition_for_local_files, data_version_by_partition_for_rclone_files) now support generating URL-encoded partition keys when encode_partition_keys is enabled.
✅ Comprehensive Test Coverage: Added extensive unit tests for partition key encoding/decoding utilities and their integration within IO managers and ops to ensure robustness and correct behavior.

Changed

🔄 Pipeline Definition Factories Updated: default_definitions, default_local_filesystem_to_datalake_definitions, and default_rclone_to_datalake_definitions now accept an encode_partition_keys parameter, providing a centralized control point for this new functionality.
⚠️ Deprecation Warning for encode_partition_keys Default: When encode_partition_keys is not explicitly set in pipeline definition factories, a deprecation warning is now emitted, indicating that the default behavior will change from False to True in a future release.

[v0.267.3] - 2026-02-26 - Enhanced User Experience with Dedicated Account Management

Added

🚀 Introduced 'My Account' API Endpoint: A new dedicated API endpoint /my-account allows users to retrieve and manage their personal profile and dashboard settings, separate from administrative user management.
✨ Created MyAccountController and Service: Implemented new backend components (MyAccountController and MyAccountService) to handle personal user account data, including profile details and dashboard configurations.
📄 Added 'My Account' Frontend Page: A new my-account.vue page provides a centralized view for users to inspect their profile, roles, last access time, and granular access permissions across services, agents, and processes.
🌐 Expanded Internationalization for Account Management: Added new translations across all supported languages for "My Account," "Services," "Agents," and "Processes" to support the new user profile section.

Changed

🔄 Refocused UserController for Administrative Tasks: The existing UserController has been narrowed in scope to exclusively manage administrative user operations within a tenant, such as listing all users or retrieving specific user details by OID.
🧹 Updated UserService to Admin-Level Operations: The UserService now focuses solely on backend logic for administrative user management, with personal account-related functions moved to the new MyAccountService.
🖼️ Revised RoleController Icon: The icon for the RoleController has been updated from mage:users to mage:security-shield to better reflect its function of managing security roles.
⚡️ Adjusted Default Landing Page for Users: Non-admin users are now automatically redirected to the /service/openai page upon login, improving the initial user experience by guiding them directly to functional areas.
📝 Clarified User Management Titles: Updated the "User" title in internationalization files to "Users" (English) and corresponding terms in other languages, to distinguish between individual user accounts and the administrative list of users.

Refactor

🏗️ Architectural Separation of User Account and User Management: Significant architectural refactoring to cleanly separate endpoints and services for a user's personal account settings (now /my-account) from administrative user management (still /users), enhancing modularity and security.
🧹 Migrated Personal Dashboard Endpoints: Dashboard retrieval and update endpoints previously under /users/me/dashboard have been migrated to the new /my-account/dashboard route, aligning with the new account management structure.

[v0.267.2] - 2026-02-26 - Streamlined Release Note Extraction

Added

✨ New extract-release-notes Makefile Target: Introduced a dedicated Makefile target to standardize and centralize the process of extracting version-specific release notes from CHANGELOG.md, improving reusability and maintainability.

Changed

🔄 Updated Release Workflow: The GitHub Actions release creation process now utilizes the new extract-release-notes Makefile target for generating changelog entries, simplifying the workflow configuration and reducing redundancy.

Refactor

🧹 Enhanced Release Note Extraction Logic: Refactored the underlying logic for extracting release notes to include robust input validation for version tags and improved fallback handling when a specific changelog section is not found.

[v0.267.1] - 2026-02-25 - Streamlined Deployment with Self-Contained Release Bundles

Added

✨ New GitHub Release Workflow: A comprehensive GitHub Actions workflow (create-release.yml) was introduced to fully automate the creation and publishing of self-contained release bundles to GitHub. This includes generating, verifying, archiving, and uploading CPU and GPU deployment packages.
🔐 Automated Environment Setup Script: A new setup-env.sh script now automatically generates secure .env files for production deployments, filling in unique, cryptographically strong secrets for database passwords, API keys, and other sensitive values from a template.
📦 generate-release Makefile Target: A new Makefile target has been added to facilitate the creation of version-pinned Docker Compose and configuration bundles for official releases, integrating seamlessly with the automated release workflow.

Changed

🚀 Fundamental Deployment Strategy: The recommended production deployment method has transitioned to downloading self-contained CPU and GPU release bundles directly from GitHub Releases, simplifying initial setup and updates significantly.
📄 Deployment & Update Documentation: The "One-Command Deployment Guide" and "Updates and Maintenance" documentation have been completely revised to reflect the new bundle-based deployment, automated environment configuration, and streamlined update processes.
⚙️ Docker Compose Generation: The underlying generate_compose.py script has been substantially upgraded to support a new "release mode" for creating deployable bundles with version-pinned image tags and simplified configuration filenames.
🛠️ Local Development Setup: Local setup instructions are now streamlined to encourage git clone of the repository and the use of a new make local-cert command for easier certificate generation.

Refactor

🧹 CI/CD Workflow Run Naming: Standardized run-name configurations across all build and deploy GitHub Actions workflows for clearer identification and tracking of CI/CD pipeline executions.
🔄 Docker Compose Template Flexibility: Refactored Docker Compose templates to introduce a config_file_suffix variable, allowing for cleaner filenames without stage or hardware specific suffixes in generated release bundles.

[v0.267.0] - 2026-02-25 - Secure Document Access and API Simplification

Added

✨ Introduced API for secure document source URLs: A new endpoint (GET /databases/{database}/namespaces/{namespace}/documents/{document_id}/url) allows authenticated users to obtain a short-lived, secure URL for downloading a document's original source file, enhancing security and control over data access.
🦾 Enhanced document lookup by namespace: Added a new method (RefDoc.by_id_and_namespace) to efficiently retrieve document references using both document ID and namespace, improving data retrieval for knowledge base operations.

Changed

🔄 Updated document download mechanism in frontend: The web application now utilizes the new backend API endpoint to fetch secure, presigned URLs for document downloads. This change centralizes file access through the backend, improving consistency and security.

Removed

🗑️ Removed direct file redirection for logged-in users: The /logged-in/redirect endpoint and its associated backend logic for file redirection have been removed, streamlining the API and consolidating file access patterns.

[v0.266.5] - 2026-02-25 - Milvus Compatibility Update

Fixed

🐛 Resolved Milvus Resource Initialization: Addressed an issue where the MilvusVectorStoreResource could fail to initialize in synchronous environments (e.g., Dagster resource creation) when using pymilvus 2.6+. This was due to pymilvus's internal reliance on an asyncio event loop during initialization, and the fix ensures a loop is present to allow for seamless and robust resource setup.

[v0.266.4] - 2026-02-25 - Port Harmonization for Streamlined Local Development

Changed

⚡️ Standardized Local Development Ports: Aligned the default development ports for key services to improve consistency and reduce potential conflicts during local development:
- The Admin UI (Frontend) now runs on http://localhost:3333 (previously 3000).
- The Dagster Orchestrator now runs on http://localhost:3000 (previously 3002).
🦾 Improved Dagster Tool Integration: Updated the mcp-server-dagster script to correctly utilize the --url argument for specifying the Dagster instance, simplifying its configuration and removing the need for an inline Python patch.
📄 Updated Documentation and Configuration: All relevant documentation, skill prerequisites, environment variables (.env.dev), and Docker Compose configurations have been updated to reflect the new port assignments for the Admin UI and Dagster.

[v0.266.3] - 2026-02-25 - Enhanced Git Push Command Management

Changed

✨ Expanded git push Command Recognition: The system now recognizes and handles a significantly broader range of git push commands, including those with advanced options such as --force-with-lease, --delete, --mirror, --all, and --tags. This enhances the system's ability to understand and process complex push operations more intelligently.
🚀 Streamlined git push Interactions: Generic and specific git push operations have been refined in their handling, potentially leading to more seamless execution and fewer prompts during typical git push workflows.

[v0.266.2] - 2026-02-25 - Streamlined Workflow Configuration

Changed

🔄 Streamlined GitHub Actions YAML: Updated the additional_permissions configuration in claude.yml to use a more concise single-line scalar format, improving the readability of workflow definitions.

Refactor

🧹 Cleaned Workflow Configurations: Removed unnecessary blank lines and commented-out placeholders from both claude-code-review.yml and claude.yml, enhancing file clarity and reducing clutter within the workflow definitions.

[v0.266.1] - 2026-02-24 - Changelog Precision Update

Fixed

🐛 Updated changelog entry date: Corrected the placeholder date for the v0.262.2 release from YYYY-MM-DD to 2026-02-17 to ensure historical accuracy.

Added

✨ Introduced Multi-Tenant Architecture: Core entities such as TenantEntity and UserTenantRoleEntity enable robust organizational isolation and tenant-scoped user management.
🔑 New Local Role Management: Roles are now managed directly within the platform's database, supporting both system-wide and tenant-specific roles, which reduces external dependencies for role synchronization.
🚀 Enabled Automatic Default Tenant Initialization: A default tenant is now automatically created and configured during the initial startup, ensuring immediate multi-tenant functionality.
👥 Implemented Configurable User Signup Roles: Roles are now automatically assigned to the first administrator and subsequent regular users upon their initial login, streamlining user onboarding.
📄 New SDK documentation for the Agent Execution Model and a detailed Event Reference guide, providing deeper insights into workflow design and event-driven patterns.
⚙️ Added Default Tenant Settings and User Signup Settings for precise control over initial tenant setup and user role assignments.

Changed

🔄 Refactored Authentication Stack: The IdentityProvider abstraction has been removed, simplifying AuthHandler implementations and consolidating identity resolution logic directly within the authentication handlers.
🛡️ Enhanced Two-Stage Access Control: The AccessChecker now implements a critical two-stage authorization process where a tenant's access rules act as a strict upper bound for all user permissions within that tenant.
⚡️ Tenant-Aware API Endpoints and Services: All relevant API routes (e.g., roles, threads, users) and core services (e.g., ThreadService, UserService, UsageLimits) have been updated to enforce multi-tenant awareness, ensuring data isolation and correct permission handling.
👤 Decoupled User Roles from UserEntity: UserEntity no longer directly stores user roles; instead, roles are dynamically fetched from UserTenantRoleEntity based on the user's active tenant context.
🤖 Integrated Bot Authentication: The AI-Hub Bot's authentication process now fully utilizes the new local user and tenant role management, ensuring secure and tenant-scoped bot interactions.
📚 Updated Platform Documentation: Extensive revisions across architectural documentation (arc42) and user guides to reflect the new multi-tenancy model, local role management, and updated authentication flows.
🎨 Minor UI Layout Adjustment: Adjusted spacing in the default layout of the web user interface for a more refined visual experience.

Fixed

🐛 Resolved an issue where UserEntity.profile_image could store invalid data URLs; it now strictly enforces valid http:// or https:// URLs for improved data integrity.

Removed

🗑️ Deprecated the IdentityProvider Abstraction: The IdentityProvider interface and all its implementations (e.g., AzureIdentityProvider, TokenIdentityProvider) have been removed to streamline the authentication architecture.
🗑️ Eliminated AUTH_IDENTITY_PROVIDER and SUPERUSER_ENABLED environment variables, as their functionality has been integrated into the new, more flexible authentication and tenant management system.
🗑️ Removed the generate_api_token.py utility script and its corresponding Makefile target, as API token generation is now managed through the refined API and aligned with the new authentication model.
🗑️ Deleted outdated RBAC SDK documentation (aihub_doc/docs/3_sdk/5_advanced_topics/5_rbac), which has been superseded by new, comprehensive multi-tenancy and access control guides.

Refactor

🧹 Upgraded Document Parsing Engine: Replaced Docling with MinerU across the platform and its documentation, providing enhanced capabilities for document processing.
🧹 Simplified Development Authentication: Streamlined DangerousDevelopmentOnlyAuthHandler by removing its dependency on IdentityProvider, reflecting the consolidated authentication architecture.
🧹 Consolidated User Onboarding Logic: Centralized user creation and default tenant role assignment into UserEntity.ensure_user_exists_for_auth for a more consistent and robust user provisioning process.
🧹 Standardized Test Fixtures: Introduced new test fixtures (mock_tenant_entity_autouse, mock_user_entity_autouse) to standardize mocking multi-tenant authentication components in tests.

[v0.265.3] - 2026-02-24 - Enhanced Service Health Monitoring for Docker Compose

Added

✨ Implemented standardized HTTP health check labels for various core services, including SeaweedFS Volume, Ferretdb, NATS Server, Presidio Analyzer, Presidio Anonymizer, and Langfuse Server. These labels provide explicit metadata (protocol, port, path) for external monitoring and orchestration tools to determine service readiness and liveness.
🚀 Configured native Docker health checks for the Jupyter Notebook service, introducing robust internal validation of its operational status with defined intervals, timeouts, and retries.
🔄 Explicitly set Docker's internal health check to NONE for Ferretdb and Langfuse Server services. This clarifies the monitoring strategy, indicating that external systems should rely on the newly added health check labels for status checks rather than Docker's default internal mechanism.

[v0.265.2] - 2026-02-24 - Comprehensive Architecture Documentation and AI-Powered Management

Added

✨ Arc42 Skill: Introduced a new Claude skill designed to write, edit, and review architecture documentation chapters following the official arc42 framework. This enhances the ability to generate structured, high-quality architectural documentation.
📖 Complete Arc42 Documentation Set: Added all 12 chapters of the arc42 architecture documentation (Introduction, Constraints, Context, Solution Strategy, Building Blocks, Runtime, Deployment, Crosscutting Concepts, Architecture Decisions, Quality Requirements, Risks, and Glossary). This provides a foundational and detailed understanding of the platform's design, operational aspects, quality goals, risks, and deployment views.

Refactor

📚 Centralized Glossary: Consolidated all project-specific terminology from individual component README.md files into a single, comprehensive arc42 glossary (Chapter 12). This significantly improves documentation consistency and ease of reference across all stakeholders.

Removed

🗑️ Redundant Component Glossaries: Eliminated duplicate glossary sections from component-specific README.md files (e.g., aihub_agent, aihub_api, aihub_bot, aihub_lib, aihub_pipeline, aihub_process, aihub_web), streamlining documentation and reducing information fragmentation.

[v0.265.1] - 2026-02-23 - Revolutionizing AI-Assisted Development with Comprehensive Claude Code Integration

Added

🦾 Extended Claude Code Integration: Introduced a comprehensive suite of new capabilities for AI-assisted development, including 7 specialized Custom Subagents (e.g., architect, codebase-expert, test-analyzer), 8 Automated Hooks for enforcing code quality, security, and environment setup, and 39 detailed Skills to streamline workflows from scaffolding to debugging across all project scopes.
🔗 New MCP Server Integrations: Integrated 11 additional Model Context Protocol (MCP) servers, providing enhanced, real-time access to documentation and runtime insights for key technologies such as Langfuse, Context7 (general library docs), Playwright, PostgreSQL, PrimeVue, Nuxt, Milvus, and NATS.
🛠️ Developer Tooling: Added new pre-commit hooks for automated Python (Ruff), YAML, and Markdown formatting, along with checks for common issues, ensuring code consistency before commits.
📄 Critical Documentation: Introduced new, in-depth documentation pages for Agent Execution Model and Events Reference, providing foundational knowledge for building robust AI agents.
✍️ Claude Code Context Files: Created dedicated CLAUDE.md context files for aihub_action, aihub_api, aihub_bot, aihub_doc, aihub_lib, aihub_process, aihub_web, and deployment scopes, offering AI assistants precise, scope-specific guidance.
✨ Developer Convenience: Added new Makefile targets in aihub_web for streamlined frontend development tasks like formatting, linting, and SDK generation.

Changed

⚙️ Claude Code Configuration Overhaul: Rewrote the core .claude/settings.json to enable granular tool permissions, define available AI models, integrate automated hooks, and activate advanced Claude Code plugins, significantly enhancing AI assistant capabilities and security.
🚀 Main CLAUDE.md and README.md: Updated the primary project documentation to reflect the extensive new Claude Code features, revised architectural overview, and updated tooling ecosystem.
📝 Agent Documentation Enhancements: Expanded existing documentation for agent fundamentals, Human-in-the-Loop (HITL), and memory integration to cover new patterns like Bot-in-the-Loop (BITL), dynamic HITL type selection, and config-driven memory management.
🔄 Backend Linting Action: Switched the lint_backend GitHub Action from using black to ruff format for Python code formatting, aligning CI with local development tooling.
🏗️ Process Endpoint Configuration: Modified the ProcessEndpointsDiscoveryService to now expect explicit name and description fields for process configurations, removing previous conditional defaults.

Fixed

🐛 Dagster UI Port Correction: Updated the documented Dagster UI access port from 3000 to 3002 in aihub_pipeline/README.md for accuracy.

Removed

🗑️ Legacy Claude Code Commands: Replaced the outdated .claude/commands/ directory with the new, more powerful and structured Claude Code Skills system.
🧹 Old Gemini CLI Integration: Removed the AGENTS.md context file and .gemini/settings.json, standardizing AI assistant context management exclusively through the .claude/ directory.
🌐 Obsolete Frontend Testing Files: Removed old placeholder frontend testing assets (index.html, logo.png, script.js) from the aihub_api playground.

Refactor

🧹 YAML File Consistency: Standardized YAML file formatting across the repository by adding the --- document start indicator to numerous configuration and i18n files.
🔧 Docker Compose Restart Syntax: Updated restart: no to restart: "no" in Docker Compose templates and generated files for improved YAML parser compatibility and consistency.
📄 Makefile Formatting Tasks: Streamlined code formatting workflows by integrating format-md and a new format-yaml target into the main pr-ready Makefile command.
🩹 MCP Script Robustness: Improved the .claude/mcp/mcp-aihub-api.sh and mcp-mongodb.sh scripts with more robust, conditional sourcing of environment variables.
📝 Gitignore Updates: Expanded .gitignore to include new Claude Code local override files, ensuring a cleaner and more focused repository.

[v0.265.0] - 2026-02-23 - Major Infrastructure Upgrade: Embracing uv for Superior Python Dependency Management

Added

✨ Introduced uv Workspaces: The project has fully migrated to uv workspaces, simplifying Python package management across the monorepo with a single virtual environment and lockfile.
📄 New Architectural Decision Record: Added a detailed Architectural Decision Record (2026_02_18_migrate_from_poetry_to_uv.md) outlining the strategic decision and benefits of migrating from Poetry to uv.
⚙️ Standardized Python Version: A new .python-version file is now included to explicitly declare Python 3.13 as the required version for the project, ensuring consistent environments.

Changed

🔄 Updated Core Development Stack: All Python package management has transitioned from Poetry to uv, drastically improving dependency resolution and installation speed for developers.
🛠️ Refined CI/CD Workflows: GitHub Actions for tagging, testing, and linting have been updated to leverage uv and its caching mechanisms for faster, more reliable, and consistent execution.
🐳 Optimized Docker Builds: Dockerfiles for all microservices (agents, API, bots, pipelines) now utilize uv for lean, multi-stage builds, eliminating Poetry from runtime images and simplifying application entrypoints.
📝 Comprehensive Documentation Updates: All developer guides, AGENTS.md files, and quick start instructions across the monorepo have been extensively updated to reflect the new uv-based setup and workflows.
💡 Improved Makefile Targets: The root Makefile now includes uv commands for setup, test, format-md, changelog, license-check, generate-compose, generate-api-token, and version-bump, streamlining the developer experience.
🏷️ Renamed License Report: The project's license report is now generated as LICENSE_REPORT.md instead of LICENSES.md for clearer identification and consistency.
🎨 IDE Configuration Alignment: Updated PyCharm/IntelliJ IDEA project and run configurations to seamlessly recognize and utilize the uv virtual environment and workspace structure.
💻 Enhanced Frontend Linting Script: The lint script in aihub_web/aihub_web/package.json has been made more robust, ensuring nuxi prepare only runs when necessary.

Fixed

🐛 Robust Process Configuration Fetching: Improved process configuration fetching logic in ProcessEndpointsDiscoveryService to ensure name and description fields correctly fall back to default values if not explicitly provided in the configuration.
🔒 Explicit Azure Identity Provider in Tests: Added an explicit environment variable (AUTH_IDENTITY_PROVIDER) in OpenWebuiAuthHandler tests to ensure the Azure Identity Provider is correctly mocked, preventing potential test flakiness.

Refactor

🧹 Consolidated Python Dependency Scanning: The license generation script (generate-license.sh) has been refactored to scan all Python dependencies from a single shared uv workspace environment, improving efficiency and accuracy.
♻️ Streamlined Internal Build Scripts: Removed the custom switch_dependencies.py script, which previously managed local versus remote dependency paths, as uv workspaces natively handle this functionality.
🧹 Unified Python Enum Usage: Replaced enum.Enum with enum.StrEnum in RcloneSourceConfig for improved type safety and clarity, aligning with modern Python best practices.
⚙️ Simplified Docker Entrypoints: The entrypoints for aihub_api and aihub_bot Docker images now directly execute gunicorn, removing an intermediate make run-prod call for better efficiency.

Removed

🗑️ Deprecated PR Agent Configuration: The .pr_agent.toml configuration file has been removed, streamlining project configuration.
🗑️ Eliminated Redundant Module Entries: Removed explicit module entries for sub-packages from the .idea/modules.xml file, as the IDE now infers them directly from the workspace.
🗑️ Removed Local/Remote Core Switching Tooling: Deleted the switch_dependencies.py script along with related Makefile targets (use-local-core, use-remote-core) and IDE run configurations, as uv provides native workspace dependency resolution.
🗑️ Removed aihub_lib Module Dependency in Pipeline IDE Config: The explicit module dependency for aihub_lib has been removed from aihub_pipeline/.idea/aihub_pipeline.iml, aligning with the new workspace structure.

[v0.264.0] - 2026-02-19 - Templates for Agents & Processes: Quick Start and Enhanced Configuration Management

Added

✨ Agent and Process Profile Templates: Introduced a new system to define and manage pre-configured templates for agent and process profiles, replacing the previous "default profile" mechanism and offering administrators multiple starting points for instance creation.
🦾 New Pre-built Agent Templates: Added a suite of practical templates for FewShotAgent (e.g., Structured Data Extractor, Support Ticket Classifier, Professional Tone Rewriter) and LLMWrappingAgent (e.g., Code Explainer, Email Drafter, Meeting Minutes Assistant) to accelerate common use cases.
🖼️ Intuitive UI for Templates and Cloning: Enhanced the web UI with a dedicated section to browse and select agent templates, and enabled cloning functionality for existing agent instances to pre-fill creation forms.
➕ Flexible Form Input Elements: Expanded the capabilities of InputText and LocaleInput form elements to support both LocaleString and plain str for placeholder, prefix, and suffix fields, offering more dynamic UI customization.
🗺️ Agent Navigation in UI: Introduced new navigation in the web UI to easily switch between "My Agents" and the new "Templates" view.

Refactor

🧹 Streamlined Agent and Process Configuration Models: Removed the agent_class and process_class fields from the AgentConfig and ProcessConfig Pydantic models, respectively, as these are now derived from the runtime context rather than being user-configurable. This improves data model clarity and type safety.
⚙️ Centralized Instance Configuration Management: Consolidated configuration normalization, validation, and metadata extraction logic into a new InstanceConfigHelper utility within the API, reducing code duplication and ensuring consistent handling across agent and process instances.
🔄 Generic UI Composable for Instance Creation: Developed a reusable useCreateInstanceForm composable for the web UI, standardizing the logic for creating both agent and process instances, including handling of class selection, dynamic form rendering, and initial data application from templates or cloned instances.
💾 Stricter Icon Requirements: Agent and Process classes now explicitly require an icon to be defined, ensuring consistent visual representation across the platform.
💬 Improved Agent Memory Initialization: AgentMemory now explicitly receives the agent_class during initialization, providing a clearer and more direct way to establish agent-specific memory contexts.

Removed

🗑️ Deprecated "Default Process Configuration": The legacy default_process_config field and its associated auto-creation mechanism have been removed from the ProcessClassDiscoveryResponseEvent and related backend logic, in favor of the new, more flexible template system.

[v0.263.1] - 2026-02-19 - Smarter AI Hub Streaming: Better Errors & Usage Warnings

Added

✨ Usage Warning Display: Implemented functionality to detect and display usage warnings (e.g., nearing API limits) received from the AI Hub service via response headers directly in the chat interface.

Changed

🌐 Localized Error Support: The Accept-Language header from the user's request is now forwarded to the AI Hub service, enabling localized error messages and responses from the backend where supported.
🚀 Enhanced Streaming Error Handling: Improved the robustness of streaming requests by allowing for early detection and more detailed handling of HTTP errors before processing the stream, providing clearer feedback to the user.
💡 Smarter API Error Message Parsing: The system now intelligently parses JSON error bodies from the AI Hub service to extract and display more specific and user-friendly error messages, improving diagnostic clarity.

Refactor

🧹 Code Readability Enhancements: Applied minor formatting and whitespace adjustments across pipeline configurations to improve overall code readability and maintainability.

[v0.263.0] - 2026-02-18 - Adopting MinerU for Superior Document Processing and Platform Architecture

Added

🚀 MinerU Integration: Replaced Docling with MinerU as the primary document parsing engine. This provides improved performance and quality for PDF and image processing, with advanced OCR, table, and formula extraction capabilities.
📄 MarkItDown Loader: Introduced a dedicated loader for Office documents (DOCX, PPTX, XLSX, Outlook messages), ensuring comprehensive support for a wider range of file types alongside MinerU.
✨ Flexible Image Handling: Implemented new utilities for managing images within parsed documents. This supports both S3-based storage with signed URLs and base64 embedding in markdown, offering greater flexibility for client-side rendering.
🏗️ Docker Network Isolation Documentation: Added detailed documentation on the platform's Docker network segmentation, enhancing security and clarifying internal service communication for better architectural understanding.
⚙️ Configurable Agent Forms: Introduced a "Form Duality Pattern" for agent configurations, allowing administrators to create and customize agent profiles via the Admin UI without requiring code changes.
🔗 S3 Filesystem Utility: Added a new create_s3_filesystem helper for consistent and easy S3 access across various services, simplifying file operations.
⚡️ New API Parsing Endpoint: Introduced /api/v1/parsing as a generic, implementation-agnostic endpoint for document conversion, adhering to the OpenWebUI external loader specification.

Changed

🔄 Document Parsing Core Logic: All internal DoclingLoader usage across the platform's document processing pipelines has been replaced with MineruLoader and MarkItDownLoader.
📝 API Endpoint and Translation Naming: The API controller group has been renamed from docling to parsing in translations to reflect the generalized document processing capabilities.
💡 Pipeline Ingestion Strategy: Updated the data ingestion pipeline to use a more flexible default_rclone_to_datalake_definitions factory, expanding support for various cloud storage providers via Rclone.
⚙️ Environment Variables: Migrated all DOCLING_* environment variables to MINERU_* for configuring the new document parsing services.
🌐 LLM Proxy Configuration: Updated LiteLLM configurations to route Visual Language Model (VLM) requests for OCR to the new MinerU VLM, leveraging either local GPU inference or partner-hosted cloud endpoints.
📚 Documentation Updates: Numerous documentation pages (e.g., solution overview, quick start, architecture) have been updated to reflect the transition from Docling to MinerU.
🛠️ API Request Body Limit: Implemented a use_limited_body dependency in the API to enforce a maximum file size for document uploads, preventing excessively large requests.
⚡️ PR Ready Build Target: Enhanced the pr-ready Makefile target to include generate-compose and license-check steps, ensuring configuration and compliance checks automatically.

Refactor

🧹 aihub_lib Dependency Management: Refactored aihub_lib dependencies across aihub_agent, aihub_api, aihub_bot, aihub_pipeline, and aihub_process to use local path references, streamlining the local development workflow.
🗄️ IDE Run Configurations: Streamlined API development and production run configurations in .idea to use Makefile targets, simplifying IDE setup for developers.
🔄 Image Path Sanitation: Improved the create_figures_folder_name utility with more robust URI and filename validation and sanitization.

Removed

🗑️ Docling Components: Completely removed the DoclingController, DoclingService, DoclingLoader, DoclingSettings, and all related DTOs and test files from the codebase.
❌ Docling Docker Services: Eliminated the docling and vllm-docling Docker services, along with their associated image tags and initialization containers.
⛔ Old API Endpoint: Deprecated and removed the /api/v1/docling API endpoint.

[v0.262.4] - 2026-02-17 - Introducing Role-Based Usage Limits and Enhanced API Feedback

Added

✨ Introduced Role-Based Usage Limits: A new powerful system enables administrators to define granular, pattern-based usage limits for agents within specific roles. These limits are enforced in real-time by the API using Redis for atomic counting.
⚡️ Usage Limit Enforcement in API Endpoints: Integrated real-time usage limit checks into both OpenAI-compatible chat completion endpoints and direct agent event endpoints, returning a HTTP 429 Too Many Requests status when limits are exceeded.
🛠️ New Usage Limit Configuration UI: Added a dedicated editor within the role management interface, allowing easy configuration and management of pattern-based usage limits for each role.
🌐 Localized Usage Limit Messages: Implemented comprehensive internationalization for usage limit warnings and exceeded messages, providing clear feedback to users in multiple languages (DE, EN, FR, IT).
📄 Structured Error Responses for Rate Limits: The API now provides detailed, structured error responses for exceeded usage limits, including localized messages and information about when the limit resets.
⚠️ Usage Warning Headers for Streaming APIs: Streaming API responses now include X-Usage-Warning headers and messages when a user's usage approaches a defined limit, enabling client applications to provide proactive notifications.
✍️ Access Rule and Usage Pattern Help Texts: Added informative help texts and examples directly within the Role editor UI to guide users in defining effective access rules and usage limit patterns.

Changed

🔄 Improved API Error Display in Web UI: Enhanced the web application's global error handler to parse structured API error responses (such as usage limit exceeded messages) and display more user-friendly details instead of raw error messages.
⬆️ Expanded Access Rule Pattern Characters: Access rule patterns now support uppercase letters, offering greater flexibility in naming conventions for resources.
⚙️ Updated OpenWebUI Pipeline for Usage Warnings: The OpenWebUI integration pipeline now processes X-Usage-Warning headers from the API to display in-chat notifications when a user is nearing their agent usage limit.
📡 Refined OpenWebUI API Error Handling: The OpenWebUI integration pipeline's streaming service now proactively captures HTTP error responses before stream processing begins, parsing structured error bodies for more informative user messages.
📜 Simplified HTTP 429 Error Message: The generic Too Many Requests error message in the web UI has been made more concise.

Refactor

🧹 Modularized Role Editor Components: The access rules editing functionality has been extracted into a dedicated AccessRulesEditor component, and a new UsageLimitsEditor component was created, significantly improving the maintainability and clarity of the role editor.
🎨 Enhanced Role Creation UI: The entry point for creating new roles in the web interface has been redesigned for a cleaner, more intuitive user experience.

[v0.262.3] - 2026-02-17 - Enhanced Developer Tooling and IDE Configurations

Added

🚀 Introduced Service Run Configurations: Added new PyCharm configurations to facilitate direct execution of core services, such as the API, from within the IDE for improved development workflow.
🛠️ Expanded Makefile Target Configurations: Integrated new PyCharm run configurations for critical development tasks, including Generate Compose, License Check, Use Local Core, and Use Remote Core, making common operations more accessible.

Changed

🔄 Renamed Test Configurations: Clarified existing PyCharm test configurations by appending "Tests" to their names (e.g., API Tests, Agent Tests, Bot Tests, Lib Tests, Process Tests), enhancing consistency and discoverability for developers.

[v0.262.2] - 2026-02-17 - Introducing the LLM-Powered Whitepaper Generation System

Added

✨ LLM-Powered Whitepaper Generation System: A comprehensive, iterative system has been introduced to automatically transform technical documentation into business-focused whitepapers, ensuring consistency, professional output, and maintainability.
📄 Automated Chapter Content and Structure: Each whitepaper chapter is now generated via an LLM, organized in self-contained folders with dedicated prompts, source mappings, and generated outputs (chapters/*/).
🔍 LLM-Based Source Discovery: Implemented generate-sources.py to intelligently identify and map relevant technical documentation files for each chapter, dynamically keeping content synchronized with evolving technical documents.
🚀 Python-Based Iterative Generation: The generate-whitepaper.py script facilitates sequential chapter generation, leveraging Jinja2 templates and previous chapters as context to maintain narrative flow and style consistency.
💰 Integrated Cost Tracking and Observability: LLM calls within the generation process now include token usage and estimated cost summaries for better financial transparency and operational oversight.
📚 Centralized Glossary and General Writing Guidelines: New configuration files (glossary.md, general_prompt.md) are used to enforce consistent terminology and writing style across all generated whitepaper chapters.
📈 Professional PDF Output with LaTeX: The system now supports direct Markdown to LaTeX to PDF conversion using pandoc and a custom LaTeX template, ensuring high-quality, professional-looking whitepapers.
🧹 Automated Markdown Formatting: Generated .md files are automatically formatted using mdformat to ensure consistent readability and adherence to project styling.
📝 Architectural Decision Record: An arc42 decision record (2025_12_05_llm_based_whitepaper_generation_system.md) outlines the context, drivers, decision, and consequences of adopting this new LLM-based whitepaper generation system.

[v0.262.1] - 2026-02-17 - Expanded External Service Configuration

Added

✨ Milvus Vector Database Configuration: Introduced new environment variables (MILVUS_URL, MILVUS_DIMENSION, MILVUS_ROOT_PASSWORD) across deployment files, enabling flexible configuration and integration with the Milvus vector database.
🔑 Docling API Key Support: Added the DOCLING_API_KEY environment variable to allow for secure authentication and access control for the Docling document processing service.

[v0.262.0] - 2026-02-12 - Comprehensive Observability Upgrade: Migrating to Langfuse for LLM Tracing and Evaluation

Added

✨ Langfuse Observability Platform: Integrated Langfuse as the new open-source LLM observability and evaluation platform, replacing Arize Phoenix. This provides enhanced tracing, cost tracking, dataset management, and UI-driven experiment workflows.
📦 Langfuse Docker Services: Introduced clickhouse, langfuse-worker, and langfuse-web services to the Docker Compose stack for robust self-hosted Langfuse deployment.
⚙️ Automated Langfuse Provisioning: Added a LangfuseProvisioner to automatically configure Langfuse on API startup, including registering AI-Hub agent models, LLM connections (e.g., LiteLLM), and default prompt templates.
🔑 Comprehensive Langfuse Configuration: New environment variables for Langfuse API keys, database settings, SSO (Azure AD) integration, and access control for production deployments.
📊 Evaluation Dataset Management API: Introduced new API endpoints under /datasets for creating, retrieving, and updating evaluation datasets in Langfuse, supporting structured testing of AI agents.
📈 Langfuse Trace Attributes: Enhanced agent tracing (AgentRunTracer) to enrich OpenTelemetry spans with Langfuse-specific trace-level attributes (name, session, user, input/output, usage details) for richer visualization and analytics.
🛡️ Increased OpenTelemetry Span Limits: Expanded the maximum number of attributes per span to 512 to prevent truncation of detailed telemetry data, especially for complex RAG traces.

Changed

🔄 Core Observability Switch: Replaced all references to Arize Phoenix with Langfuse across the entire platform, including documentation, code comments, and configuration files, for a consistent observability experience.
🔬 Streamlined LLM Evaluation Workflow: Shifted from a custom, programmatic experiment evaluation framework to leveraging Langfuse's native UI for managing and running experiments against datasets, simplifying evaluation processes and reducing custom code.
📡 Agent Instance Sync to Langfuse: The AgentEndpointsDiscoveryService now automatically syncs online agent instances to Langfuse, ensuring they are visible and selectable for experiment evaluation within the Langfuse UI.
📄 Updated Documentation: All relevant documentation has been updated to reflect the transition to Langfuse, detailing its features and usage for tracing and evaluation.

Refactor

🧹 Simplified AgentRunTracer Logic: Refactored the AgentRunTracer to directly leverage Langfuse's OpenTelemetry ingestion capabilities, removing Phoenix-specific logic for root spans and complex context storage.

Removed

🗑️ Arize Phoenix Components: Eliminated all Docker services, configuration, API endpoints, and associated Python code (including PhoenixExperimentEvaluator and JudgeOutput) related to Arize Phoenix due to licensing incompatibility and technical advantages of Langfuse.
🚫 Experiment Management Frontend: Removed the custom frontend components for managing and running evaluation experiments, as these functionalities are now handled directly within the Langfuse UI.

[v0.261.7] - 2026-02-11 - Enhanced Docling Configuration with API Key Support

Added

🔑 Docling API Key Support: Introduced a new configuration option (DOCLING_API_KEY) to allow specifying an API key for the Docling service, improving secure access and authentication.

[v0.261.6] - 2026-02-11 - Enhanced Deployment Flexibility and Routing Accuracy

Added

🚀 Introduced dynamic frontend configuration: Environment variables (AIHUB_API_VERSION, AIHUB_FRONTEND_ORIGIN) are now passed to AI-Hub frontend services in Docker Compose setups, enabling more flexible and dynamic deployment settings.

Changed

⚙️ Optimized Nuxt.js prerendering: Specific authentication routes (e.g., /en/auth) are now explicitly ignored during Nuxt.js prerendering, which enhances build efficiency and better accommodates dynamic authentication pages.

Fixed

🐛 Resolved authentication path matching issues: The authentication middleware now correctly identifies public paths, regardless of trailing slashes, ensuring consistent access to non-authenticated routes.
🐞 Corrected Nginx absolute redirect behavior: Disabled absolute redirects in Nginx to prevent issues with incorrect schemes or ports when the application is deployed behind a reverse proxy.
⚡️ Improved Nginx SPA routing: Refined the Nginx configuration for single-page applications to ensure all non-file requests are reliably served by index.html.

[v0.261.5] - 2026-02-10 - Streamlined Development and Dependency Management

Refactor

🔄 Monorepo Dependency Alignment: Updated the aihub_lib dependency across aihub_agent, aihub_api, aihub_bot, aihub_pipeline, and aihub_process to use local path references with develop = true. This change streamlines local development workflows and better supports a monorepo project structure.

Removed

🗑️ flatdict Dependency: The flatdict library has been removed from the aihub_agent package's dependencies and its documentation, simplifying the project's dependency footprint.

[v0.261.4] - 2026-02-10 - Internal Dependency Streamlining

Refactor

🧹 Streamlined event attribute flattening: Replaced the external flatdict library with a custom, in-house implementation. This change reduces external dependencies and improves control over how event attributes are processed and presented for display.

[v0.261.3] - 2026-02-09 - Dependency Script Logging Enhancements

Changed

📄 Enhanced poetry Command Output: The switch_dependencies.py script now prints the standard output (stdout) from poetry lock and poetry install commands, offering more transparency into dependency management processes.
🚨 Clarified Dependency Installation Errors: Error messages generated by switch_dependencies.py for failed poetry lock and poetry install operations have been improved to explicitly display the standard error (stderr) only when relevant, preventing empty or unhelpful error outputs.

[v0.261.2] - 2026-02-09 - Streamlined CI/CD Builds with Dynamic Image Discovery

Refactor

🔄 Dynamic Agent Image Discovery: Improved the build-agents.yml workflow to automatically determine which agent images to build by parsing compose-config.yml, enhancing build system flexibility and reducing manual configuration.
🔄 Dynamic Pipeline Image Discovery: Enhanced the build-pipelines.yml workflow to dynamically identify pipeline images for building directly from compose-config.yml, streamlining the CI/CD process and ensuring consistency.

[v0.261.1] - 2026-02-09 - Comprehensive Changelog History Update

Added

📄 Comprehensive Changelog History: Updated CHANGELOG.md to include detailed release notes for recent versions, specifically adding entries for v0.261.0, v0.260.2, v0.254.6, v0.254.5, v0.243.3, and v0.243.2, ensuring a complete and up-to-date historical record of changes.

[v0.261.0] - 2026-02-06 - Dynamic Configuration and Enhanced Management for Agents & Processes

Added

⚙️ Dynamic Agent Configuration via Admin UI: Introduced the Form Duality Pattern enabling developers to define agent configurations directly in code that are automatically rendered as editable forms in the Admin UI. This empowers administrators to customize agent behavior without code changes.
✨ New Agent Profile Management API: Implemented a comprehensive set of API endpoints for managing agent instances (profiles) including creation, retrieval, updates, and deletion. This replaces previous discovery-only mechanisms with robust CRUD operations.
✨ New Process Instance Management API: Similarly, introduced a complete API for managing process instances, offering full control over process deployment and configuration through the Admin UI.
⚙️ NATS RPC for Dynamic Configuration Retrieval: Added a new NATS RPC mechanism (AgentConfigClient, ProcessConfigClient, AgentConfigResponder, ProcessConfigResponder) enabling agents and processes to fetch their runtime configurations dynamically from the API, decoupling configuration from event payloads.
✨ New FormKit UI Elements: Developed custom FormKit components (AgentSelector, IconSelector, KnowledgeDatabaseSelector, LocaleInput, ModelSelect, Repeater, VectorStoreInput) to support rich, dynamic and multi-language configuration forms in the UI.
📝 Comprehensive Configuration Documentation: Added detailed documentation (arc42/decisions/2026_01_07_enable_dynamic_agent_configuration_ui.md, docs/2_platform/5_agents/3_blueprints_and_profiles, docs/3_sdk/2_building_agents/8_configurable_agents) explaining the new configuration paradigm, agent blueprints vs. profiles, and how to build configurable agents.
⚙️ New Agent Implementations: Moved and refactored FewShotAgent, RetrievalAgent, and NamespaceSelectionAgent from playground to production-ready app directories, making them configurable via the new UI.
⚙️ LLM-Based Translation Service: Introduced a new API route and service for performing LLM-based translations of LocaleString objects across all supported languages (de, en, fr, it).
✨ Agent and Process Class Metadata: Added class-level name, description, and icon properties to Agent and AgenticProcess base classes, enabling consistent display of agent/process types in the UI.
⚙️ New Makefile Targets: Added run-dev, format-md, and pr-ready targets for local development, markdown formatting, and pull request preparation.
✅ Expanded Test Coverage: Introduced new unit tests for AgentConfig, Form classes, and NATS RPC mechanisms, ensuring the robustness of new features.
⚙️ Enhanced OpenWebUI Integration: Added a new OpenWebUI function (memory_action) to allow viewing agent memories directly from the AI-Hub frontend.

Changed

🔄 API Restructuring for Agents and Processes: Overhauled the /agents and /processes API endpoints to reflect the "class vs. instance" (blueprint vs. profile) distinction, providing more granular control and a clearer hierarchy for management.
🔄 Centralized Configuration Fetching: Refactored AgentService and ProcessService to adopt a database-first approach for configuration. Agent/Process dispatchers now fetch runtime configurations dynamically via NATS RPC from the API service, rather than relying on event payloads or local defaults.
🔄 Updated AgentRunner and ProcessRunner: Modified runners to use the new form-mode configuration instances and expose class-level metadata for discovery.
🔄 Refined StartEvent Structure: Removed the agent_config field from StartEvent and UserMessageEvent, streamlining event payloads as configuration is now fetched dynamically by the receiving agent/process.
🔄 Dependency Updates: Updated numerous dependencies in aihub_web to their latest versions, including @pinia/colada, @vueuse, gridstack, oidc-client-ts, and socket.io-client, improving performance and stability.
🔄 Improved UI/UX Icons: Updated various icons across the web frontend (e.g., chat, file, search, agent status) for a more consistent and modern look and feel.
🔄 Console Logging Level Adjustment: Changed the default logging level for several external libraries in aihub_lib to ERROR to reduce noise in console output.
🔄 Unified Workflow Visualization: Updated WorkflowVisualizer to use localized strings for start/end nodes, improving multi-language support.
🔄 Configurable Milvus Vector Store: Refactored MilvusVectorStoreConfig to expose index_namespaces for explicit configuration and now fetches connection details from MilvusSettings at runtime.
🔄 Refined OpenWebUI Integration: Updated existing OpenWebUI integration to align with the new API endpoint structure for agents and dynamic display of agent names.

Fixed

🐛 UI Console Log Duplication: Fixed an issue where console logs were excessively verbose by adjusting default logging levels for common libraries.
🐛 Process Page Title Display: Corrected a typo in the process details page title.
🐛 Agent StartEvent Test Data: Updated test event data to correctly include agent ID, aligning with new event structures.

Removed

🗑️ Deprecated WebuiAgent: Removed the WebuiAgent and all associated files, streamlining the agent portfolio.
🗑️ Obsolete open_webui SDK: Removed the internal OpenWebUI Python SDK as it is no longer actively maintained.
🗑️ Legacy Agent and Process Discovery Caches: Eliminated in-memory caches (e.g., DISCOVER_AGENTS_CACHE, GET_AGENT_INSTANCE_CACHE) as discovery and configuration now leverage a database-first approach and NATS RPC.
🗑️ Redundant Agent and Process Entities: Removed AgentEntity and ProcessEntity database models, replacing them with AgentClassEntity and ProcessClassEntity for class-level data, and AgentConfigEntityDocument for instance-specific configurations.
🗑️ Outdated Playground Run Scripts: Removed various run.py and trigger.py scripts from the playground directories, as agents have been migrated to the app directory and integrated with the new configuration system.
🗑️ Unused Agent/Process Instance Discovery Events: Removed AgentInstanceDiscoveryRequestEvent, AgentInstanceDiscoveryResponseEvent, ProcessInstanceDiscoveryRequestEvent, and ProcessInstanceDiscoveryResponseEvent, shifting to class-level discovery.

Refactor

🧹 Unified Configuration Logic (Form Class): Consolidated form generation, data validation, and configuration handling into a single Form base class in aihub_lib, enabling a consistent "duality pattern" for defining UI-editable configurations across agents and processes.
🧹 Centralized Type Hinting: Applied Self type hints across numerous methods in aihub_lib, aihub_api, and aihub_bot for improved type inference and code consistency.
🧹 Cleaned Up Internal Comments: Removed extraneous comments from various core files to improve code readability and maintainability.
🧹 Streamlined .idea/runConfigurations: Reorganized and renamed numerous IntelliJ IDEA run configurations for better clarity and ease of use.
🧹 Modularized API Controllers: Refactored API controllers by extracting ApiLocaleString for localized names and descriptions, enhancing modularity and i18n support.
🧹 Normalized Form Data: Introduced new utility functions (normalize_empty_objects_to_none, normalize_empty_locale_strings, transform_formkit_arrays) to standardize form data sent from the frontend to backend Pydantic models.

[v0.260.2] - 2026-02-03 - Universal Data Ingestion with Rclone and Smarter Memory Retrieval

Added

🚀 Introduced Rclone Integration for Universal Data Ingestion: A new, unified pipeline now supports syncing documents from over 70 cloud storage providers (including SharePoint, OneDrive, Google Drive, AWS S3, Azure Blob, SFTP, and local filesystems) to your S3 data lake. This replaces previous provider-specific implementations, offering a single, flexible solution.
⚙️ New Rclone Docker Service: A dedicated rclone service has been added to the Docker Compose setup, providing the Rclone Remote Control (RC) API for seamless integration with ingestion pipelines.
📄 Comprehensive Source Templates: Pre-configured templates and detailed setup guides have been added for common Rclone sources, simplifying configuration for Azure Blob, Google Drive, Local Filesystem, OneDrive, S3, SFTP, and SharePoint.
✨ Enhanced Memory Retrieval Display: New event handlers (RetrieveUserMemoryEventHandler, RetrieveOrganizationMemoryEventHandler) improve the display of retrieved user and organization memories, providing more detailed source information in the UI.
🔑 Rclone RC API Environment Variables: Added RCLONE_URL, RCLONE_RC_USER, and RCLONE_RC_PASS environment variables for configuring the Rclone RC API connection, including secure authentication for production.

Changed

🔄 Migrated Data Ingestion Documentation: The data ingestion pipeline documentation has been comprehensively updated to reflect the new Rclone-based universal approach, deprecating the previous SharePoint-specific guide.
🛠️ Updated Data Lake Definitions Utility: The default_sharepoint_to_datalake_definitions factory has been replaced by default_rclone_to_datalake_definitions, streamlining the creation of data ingestion pipelines.

Removed

🗑️ Deprecated SharePoint-specific Data Lake Definition: The default_sharepoint_to_datalake_definitions factory has been removed, superseded by the more versatile Rclone integration.

Security

🔒 Rclone RC API Authentication Enforced: The Rclone service now requires authentication via RCLONE_RC_USER and RCLONE_RC_PASS in production environments, significantly improving security. Users are explicitly warned to change default credentials in production.

Refactor

🧹 Flexible Environment Settings Configuration: The create_settings_config method in EnvironmentSettings now supports an extra argument, allowing for more flexible loading of dynamic, backend-specific Rclone options from environment variables.

[v0.260.1] - 2026-02-03 - API Configuration Standardization for Docling Loader

Refactor

🧹 Standardized API Base URL Configuration: Renamed the internal configuration variable for the Docling API base URL from BASE_API_URL to API_BASE_URL within the DoclingLoader for improved consistency and clarity in the codebase.

[v0.260.0] - 2026-02-02 - Enhanced Memory Management and Agent Context

Added

🧠 Introduced Comprehensive User and Organization Memory: A new, robust memory system has been implemented across RAGAgent and ExpertRAGAgent, allowing agents to retrieve personalized user memories and organization-wide knowledge.
🦾 New Agent Workflow Steps for Memory Management: Added dedicated agent steps for ExpertRAGAgent and RAGAgent to retrieve user memories, retrieve organization memories, inject them into chat history, and store new user memories from conversations.
⚙️ Configurable Memory Options for RAG Agents: RAGAgentConfig and ExpertAskingAgentConfig now include explicit configuration fields to enable/disable organization memory retrieval, user memory retrieval, and user memory storage, along with tenant_id and tenant_namespace for scoping.
🗄️ Memory Management Pages in Web UI: Dedicated UI pages are now available for managing user memories associated with specific agents and threads, providing a centralized view for collected knowledge.
📊 OpenWebUI Integration for Agent Memories: Introduced new UI components and an OpenWebUI function to display agent-specific user memories directly within the chat interface, with toggleable list and graph views.
🔍 Advanced Memory Search Filters in API: Added agent_class and agent_id parameters to the User and Organization Memory search API endpoints, enabling more precise filtering of stored memories.
💬 Localized Memory Interface: Extended translation files to support the new memory management sections and views across multiple languages.
⚡️ Granular LLM Event Handling: The do_respond_with_llm step function now supports returning an LLMEvent for intermediate steps, allowing for post-LLM processing (like memory storage) before a final StopEvent is emitted.

Changed

🔄 Expert Conversation Storage Refactored: ExpertAskingAgent now stores expert conversations as general "organization memories" instead of the specialized "insights," aligning with the new unified memory system.
🧪 RAG Agent Test Suite Updated: The RAG Agent test suite has been updated to remove references to the deprecated insight system and now tests the new organization memory retrieval.
💻 SDK Schema Generalization: The Usage schema in the SDK has been generalized to represent duration, with specific token usage details moved to an image-response-specific schema.
🎨 Enhanced Memory List UI: Improved the visual presentation of the memory list component in the web UI with striped rows, better column styling, and severity-based tagging for relevance scores.

Removed

🗑️ Deprecated Insight System: The entire "insight" persistence and retrieval system, including InsightEntity, InsightRetriever, related configurations, API endpoints, and UI logic, has been removed.
🧹 Cleanup of Insight-Related Code: All code paths, imports, and translation files associated with the old insight system have been meticulously removed from the codebase.

Fixed

🐛 Improved Mem0 Metadata Filtering: Corrected Mem0 service logic to accurately filter out empty strings in addition to None values when processing metadata and filters, preventing unintended memory retrieval behavior.

[v0.259.3] - 2024-07-29 - Automated Latest Tagging Workflow Enhancement

Changed

🚀 Automated Docker Image Discovery for 'latest' Tagging: The GitHub Actions workflow responsible for promoting release tags to "latest" now dynamically identifies all custom application Docker images. It parses deployment/compose-config.yml to automatically detect images marked for local builds, removing the need for manual updates to the workflow when new services are added or removed.
📄 Improved Workflow Documentation: Comprehensive comments have been added to the set-latest.yml GitHub Actions workflow, clearly detailing its purpose, usage, and the logic used for image selection and promotion.

[v0.259.2] - 2026-01-30 - Strengthened Security with Docker Network Isolation

Security

🔑 Implemented a comprehensive Docker network isolation strategy, segmenting the platform into proxy, backend, data, storage, and egress networks to significantly enhance security posture. This limits lateral movement and reduces the blast radius in case of a breach.
🔒 Configured internal: true for core backend, data, and storage Docker networks (in non-development environments), preventing direct external access to sensitive internal services.
🚫 Explicitly disabled Inter-Container Communication (ICC) on the dedicated egress network, ensuring that services requiring outbound internet access cannot communicate with other containers on that network, further isolating them.

Refactor

🔄 Systematically updated Docker Compose deployment configurations across all environments (dev, build, local, latest, nightly) to integrate the new multi-network architecture, assigning each service to its specific, isolated network zone.

Added

📄 Introduced detailed documentation for Docker Network Isolation, providing a clear overview of the new network topology, service assignments, security benefits, and operational guidelines.
🩺 Implemented a new health check for the Open WebUI service to improve monitoring and ensure its availability.

Changed

🛠️ Migrated the Playwright service to a custom Dockerfile build process, providing greater control over its runtime environment and dependencies.
⬇️ Pinned the Playwright image version to v1.49.0-jammy to ensure stability and compatibility within the new custom build setup.
🔗 Updated service dependencies for api and bot, ensuring they now explicitly await the healthy state of Milvus and Neo4j, respectively, for more robust startup sequences.
🧹 Refined health check commands for postgres-ferretdb and litellm for improved accuracy and reliability.

[v0.259.1] - 2026-01-30 - Enhanced OAuth Security and Granular Access Control

Security

🔑 Improved OAuth Cookie Secret Management: Replaced the single shared OAuth cookie secret with individual, service-specific secrets (OAUTH_COOKIE_SECRET_DAGSTER, OAUTH_COOKIE_SECRET_SEAWEEDFS, OAUTH_COOKIE_SECRET_ATTU). This significantly enhances security by isolating session cookies for different platform components (Dagster, SeaweedFS, Attu).

Changed

🔄 Refined OAuth Group Restrictions: Renamed and segregated OAuth allowed group configurations to be service-specific (e.g., DAGSTER_OAUTH_ALLOWED_GROUPS is now OAUTH_ALLOWED_GROUPS_DAGSTER). This provides more granular control over user access for each component.
📄 Updated Environment Variable Configuration: Adjusted .env templates and all Docker Compose configurations to reflect the new, service-specific OAuth cookie secrets and allowed group variables, streamlining secure setup.
📚 Revised Documentation for OAuth Setup: Updated quick start guides and secret generation instructions to provide clear guidance on configuring the new, separate OAuth cookie secrets for each service.

Removed

🗑️ Deprecated Shared OAuth Cookie Secret: The generic OAUTH_COOKIE_SECRET environment variable has been removed in favor of service-specific secrets, reducing potential security exposure.

[v0.259.0] - 2026-01-28 - Empowering Agents with Long-Term Memory and Knowledge Graphs

Added

✨ Introduced Agent Memory System: Agents can now learn from past conversations and organizational facts, providing long-term personalization and shared knowledge, powered by mem0 and Neo4j.
🧠 User Memory Capabilities: Agents can automatically learn and store private user preferences, working styles, and individual context from conversations, enabling highly personalized interactions.
🏢 Organization Memory for Shared Knowledge: Agents can now leverage explicit, shared organizational facts (policies, tech stack, processes) accessible to all users within a tenant, fostering consistent and institutional knowledge.
📈 New Memory Management API Endpoints: Introduced comprehensive API endpoints for viewing, searching, updating, and deleting both user and organization memories via the platform's API.
📊 Interactive Memory Knowledge Graphs: The web interface now visualizes memory relationships as interactive knowledge graphs, helping users understand how concepts are connected within their memories.
📺 Dedicated Memory Management UI: New sections in the web interface provide dedicated pages for managing and exploring personal user memories and shared organization memories.
⚙️ Integrated Neo4j Graph Database: Incorporated Neo4j for powerful graph-based memory storage and retrieval, enabling rich relational understanding between concepts.
🧩 Microsoft Teams Support for Expert Asking Agent: The Expert Asking Agent can now escalate questions to human experts via Microsoft Teams, expanding its capabilities beyond Slack integration.
⏱️ Run Duration and Cost Metrics in Event List: The event list now displays the duration and total cost for each agent run, enhancing observability and cost transparency.
📄 Comprehensive Agent Memory Documentation: Added detailed guides on User Memory, Organization Memory, and how to build memory-enhanced agents using the SDK.
⚖️ Automated Python Dependency License Scanning: Python packages now have their licenses automatically scanned and reported in LICENSES.md, significantly improving license transparency and compliance.

Changed

🛠️ Refactored AI Utility Modules: Reorganized general AI utility functions into more logical and specialized modules: chat_history, rerank, and retrieval, improving code structure and maintainability.
📦 Updated LiteLLM Proxy Configuration: Improved model parameter handling in LiteLLM by adding drop_params for better compatibility and efficiency across various models.
📏 Adjusted Milvus Embedding Dimension: Updated the default Milvus embedding dimension to 1024 in development environments, optimizing vector storage for newer embedding models.
📄 Renumbered Platform Documentation Chapters: Reorganized the platform documentation structure to logically integrate the new memory features and improve navigability.
📄 Renumbered Agent SDK Documentation: Restructured the Agent SDK documentation to seamlessly integrate new memory patterns and improve readability for developers.

Fixed

🐛 Corrected Redis Host in LiteLLM Configuration: Ensured LiteLLM correctly references valkey (the Redis-compatible service) within the Docker network instead of localhost, resolving connectivity issues.

Refactor

🧹 Standardized Enum Usage: Converted several enum.Enum implementations to enum.StrEnum across the codebase for improved type safety and consistency.

Removed

🗑️ Deprecated GDPR Documentation: Removed previous GDPR documentation pages, as the content has been updated and integrated into the new compliance and memory management sections.

[v0.258.4] - 2026-01-19 - Enhanced Service Health, API Stability, and Richer Chat Interactions

Added

✨ Comprehensive Service Health Checks: Introduced new liveness and readiness endpoints for the API, Agents, and Processes, providing detailed dependency status for critical infrastructure like NATS, MongoDB, Redis, Milvus, and S3.
⚙️ Centralized Health Check Server: Implemented a generic HTTP health check server running in a background thread for Agents and Processes, ensuring consistent and robust monitoring across all services.
🚀 New Model Management API Endpoints: Added dedicated API endpoints to retrieve lists of all available models grouped by type, and to fetch detailed information for specific models.
💬 Expanded Chat Message Blocks: Introduced support for new message content types, including VideoBlock, ThinkingBlock, and ToolCallBlock, significantly enriching conversational capabilities and allowing for more complex multimodal interactions.
🗣️ Chat-style Human-in-the-Loop (HITL): Added a new chat type for Human-in-the-Loop requests, enabling users to provide input and respond to prompts directly within the regular chat interface, improving user experience.
🧩 FastAPI Infrastructure Client Dependencies: Introduced new FastAPI dependencies (use_milvus, use_redis, use_s3, use_s3_public, use_s3_service, use_vector_store_factory) to standardize access to shared Milvus, Redis, and S3 clients and vector store factories, promoting consistency and reusability.

Changed

⬆️ Agent and Process Runner Upgrades: Integrated Agent and Process runners with the newly introduced health check server, now providing detailed readiness statuses for their respective dependencies (NATS, Redis, Milvus).
✅ API Readiness Checks: The API service now utilizes the ApiHealthController for comprehensive readiness checks, moving beyond a simple liveness probe to verify all critical service dependencies.
🔗 Milvus Client Reusability: Refactored Milvus vector store creation to reuse a shared MilvusClient instance across the application, improving connection pooling efficiency and enabling centralized health monitoring.
📍 Bot Service Port Update: Corrected the exposed port for the Bot service from 8000 to 8001 to align with standard deployment configurations.
🔒 Milvus Authentication Variable: Standardized Milvus client authentication across various services by updating the MILVUS_TOKEN environment variable to MILVUS_ROOT_PASSWORD for consistency and clarity.
🛡️ Enhanced SDK Endpoint Security: Applied security headers to all endpoints within the generated TypeScript SDK client, improving overall API security.

Refactor

🧹 S3 File Access Service Modernization: Refactored the S3AnonymousFileAccessService to accept injected S3 client instances, significantly improving testability and streamlining resource management.
♻️ Knowledge Controller Dependency Injection: Migrated direct instantiation of VectorStoreFactory and S3AnonymousFileAccessService within the KnowledgeController to FastAPI's dependency injection system, enhancing modularity and maintainability.

[v0.258.3] - 2026-01-19 - Hardened Docker Integration with Socket Proxy

Security

🔑 Implemented Docker Socket Proxy: Introduced a new docker-socket-proxy service that acts as a secure intermediary between Traefik and the Docker daemon. This significantly enhances system security by limiting Traefik's access to only the essential Docker API endpoints (container and network discovery) needed for service routing, mitigating potential container escape vulnerabilities.

Added

✨ docker-socket-proxy Service: A dedicated tecnativa/docker-socket-proxy service has been added to all Docker Compose configurations, providing fine-grained control over Docker API access.
⚙️ Docker Socket Proxy Configuration: Comprehensive configurations for the new proxy service are included, explicitly disabling dangerous Docker API permissions (e.g., build, commit, exec, images) and ensuring read-only access to the underlying Docker socket.

Changed

🔄 Traefik Docker Endpoint: Traefik's Docker provider endpoint has been updated to connect to the new docker-socket-proxy over TCP (tcp://docker-socket-proxy:2375) instead of directly accessing the Unix Docker socket.
🧹 Removed Direct Docker Socket Mount: The direct volume mount of /var/run/docker.sock from the Traefik service has been removed across all deployment configurations, delegating Docker socket interaction to the new, more secure proxy service.

[v0.258.2] - 2026-01-13 - Deepening Knowledge: Insights Now Integral to Document & Node Retrieval

Added

✨ Enhanced Knowledge Base with Insight Integration: Introduced the capability to retrieve Insights directly as standard documents and nodes within the knowledge base, enabling their use in Retrieval-Augmented Generation (RAG) flows.
📚 Direct Insight Retrieval by ID: Added a new method (InsightEntity.get_by_id) to directly fetch insights using their unique identifier, streamlining access to structured knowledge.
🌍 Localized Insight Representation: Implemented support for localizing insight content when presented as nodes, ensuring that questions, answers, and conversation parts are displayed in the user's preferred language.

Changed

🔄 Unified Document and Node Retrieval: Modified the KnowledgeService to transparently handle both regular documents and Insights when fetching documents by ID or retrieving specific nodes, providing a more consistent API experience.
🛠️ Refined Insight-to-Document Conversion: Updated DocumentDTO to include a dedicated method for converting an InsightEntity into a document data transfer object, standardizing how insights are represented.
🌐 Improved Insight Source Handling in UI: Adjusted the web interface to recognize insights as a distinct document source, ensuring the "View Original" action is correctly hidden for insight-based knowledge.
🗄️ Virtual Bucket Assignment for Insights: Updated the extractBucket utility in the UI to correctly assign a virtual 'insights' bucket for documents originating from MongoDB-stored insights.

Refactor

🧹 Encapsulated Insight Node Conversion Logic: Moved the logic for transforming an InsightEntity into an IngestedNode directly into the InsightEntity class, improving modularity and reusability.
📄 Minor Logging Message Formatting: Adjusted a logging message in the DoclingLoader for improved clarity.

[v0.258.1] - 2026-01-13 - Knowledge Base Evolution: Advanced Document Management and Visibility

Added

✨ Introduced Document Search and Sorting: Users can now easily find documents by title or filename and sort them by creation date, update date, or title directly in the Knowledge Base UI.
🚀 Real-time Document Ingestion Tracking: Documents uploaded to the data lake are now immediately visible in the UI with a "pending" status, providing real-time feedback on ingestion progress.
📄 Placeholder Document Support: The backend now creates RefDoc placeholders for newly uploaded files, enabling immediate visibility and consistent tracking throughout the ingestion lifecycle.
🔗 Deterministic Document IDs: Implemented a new utility for generating deterministic IDs for RefDoc entries based on their source path, improving consistency and data management.
🧩 Dagster Pipeline Assets for Placeholders: Integrated new Dagster assets and operations to automate the creation of placeholder RefDoc entries for files entering the data lake via SharePoint and local filesystem pipelines.

Changed

🔄 Unified Document Retrieval Logic: The Knowledge Base backend has been revamped to use a single, unified approach for fetching documents from the RefDoc store, simplifying the architecture and improving consistency.
⬆️ Enhanced Document Upload Validation: The document upload process now reliably creates a RefDoc placeholder for pending files and includes improved error handling for event publishing, making uploads more robust.
📝 Clarified DocumentDTO fields: The DocumentDTO fields source and is_ingested have clearer descriptions, better reflecting the document's state and origin.
🧹 Refined Document List UI: The document list now uses consistent field names (document_title, created_at, updated_at) aligning with backend changes and supports the new sorting capabilities.
⚡️ Improved UI Loading States: The document list now provides a clearer loading indicator, only showing a full loading state on initial data fetch for a smoother user experience.
⚠️ Conditional Document Upload Button: The document upload button is now only visible for knowledge bases that are not configured for automatic synchronization, preventing manual uploads to auto-synced sources.
🗑️ Enhanced Data Lake File Deletion: Deleting files from the data lake now automatically cleans up associated RefDoc entries from the document store, ensuring data consistency.
⚙️ Human-in-the-Loop Event Structure: HumanInTheLoop request and response events have been refined to include a hitl_type and support more flexible response data types, making them more adaptable to various human interaction patterns.

Refactor

🧹 Centralized MongoDB Connection Management: MongoDB connection handling across aihub_lib and aihub_pipeline has been refactored to use switch_db context managers and dedicated helper functions, improving multi-database support and connection stability.

Removed

🗑️ Deprecated Document Caching Logic: Custom caching mechanisms for distinguishing between processed and unprocessed documents in the Knowledge Service have been removed, replaced by the new unified RefDoc retrieval strategy.
📚 Streamlined Models API: Several deprecated model-related DTOs and API endpoints have been removed or consolidated, simplifying the API surface for model management.

[v0.258.0] - 2026-01-12 - Fortified Core Services: Enhanced Security, Streamlined Configuration, and Improved Observability

Security

🔑 Introduced comprehensive token-based authentication for key infrastructure components including Milvus, NATS, Redis/Valkey, Etcd, and SeaweedFS, significantly enhancing the platform's security posture.
🔑 Enabled authentication for Docling API access, allowing API key-based security for document processing services.
🔐 Added etcd-init service to automatically set up root password authentication for Etcd on startup, ensuring secure metadata storage from the outset.
🔑 Centralized management of API keys for local Large Language Models (LLMs) (llama.cpp and vLLM) under the LOCAL_LLM_TOKEN environment variable for easier and more secure configuration.

Added

📦 Enabled Docker secrets support for Pydantic settings, allowing sensitive configuration data to be managed securely as Docker secrets, with environment variables taking precedence for development flexibility.
⚡️ Integrated LiteLLM caching with Redis, now supporting authentication to improve performance and secure cached responses.
📊 Introduced debug exporters for OpenTelemetry logs and metrics in development environments, providing immediate visibility into service telemetry.

Changed

🔄 Standardized NATS and Redis client initialization by centralizing their creation within dedicated NatsSettings and RedisSettings classes, simplifying client setup across all services (agents, API, pipelines).
📈 Activated OpenTelemetry by default in development environments to provide immediate observability into service operations.
🏷️ Implemented dynamic service versioning for OpenTelemetry resources, automatically deriving service versions from Docker image tags for more accurate telemetry data.
⚙️ Standardized Docling API configuration variables (DOCLING_BASE_API_URL to DOCLING_API_BASE_URL and DOCLING_HOSTED_VLM_API_ENDPOINT to DOCLING_HOSTED_VLM_API_BASE_URL) for consistency.
🌐 Renamed Swiss LLM Cloud API URL to SWISS_LLM_CLOUD_API_BASE_URL for improved clarity and consistency in configuration.
🚫 Disabled OpenTelemetry for local Dagster runs by default in the Makefile to prevent potential conflicts and streamline local development workflows.
🔗 Updated OpenWebUI connections to Redis and Milvus to properly utilize the newly introduced authentication tokens.
👷 Adjusted CI workflow to always rebuild Docker images for backend tests instead of relying on cached images, ensuring tests run against the freshest code.

Refactor

🧹 Migrated SeaweedFS Filer configuration entirely from static TOML files to environment variables, streamlining deployment and configuration management.
⚙️ Simplified Agent and Process runners by removing direct NATS server list and Redis URL parameters, now leveraging NatsSettings and RedisSettings for configuration.

Changed

🦾 Human-in-the-Loop (HITL) Context: Enhanced the Human-in-the-Loop response mechanism to explicitly pass agent class and ID, providing more precise context for agent interactions.

Refactor

🧹 Code Readability: Improved code readability and conciseness across various AI Hub pipeline components by optimizing variable assignments and method signatures.

[v0.257.1] - 2026-01-09 - Enhanced Knowledge Retriever Namespace Flexibility

Changed

🔄 Improved KnowledgeRetrieverConfig for Namespaces: Relaxed the constraint on index_namespaces, allowing it to be an empty list. An empty list now signifies the intention to retrieve information from all available namespaces, offering greater flexibility.
🦾 Smarter Retriever Filtering: Updated the filter_retrievers_by_namespace utility to correctly interpret an empty index_namespaces list, ensuring that retrievers are considered for all relevant namespaces when no specific namespace is provided.
⚡️ Dynamic Node Retrieval: The retrieve_nodes function now intelligently constructs metadata filters. If index_namespaces is empty, it will skip namespace-specific filtering, allowing retrieval of nodes across all namespaces based solely on node type.

[v0.257.0] - 2026-01-08 - Empowering RAG with Dynamic Namespace Selection and Enhanced Retrieval

Added

✨ Namespace Selection Agent: Introduced a new agent dedicated to interactively guiding users through the selection of relevant knowledge source namespaces for their queries. This agent uses an LLM to understand intent and incorporates human-in-the-loop (HITL) steps for clarification and approval, ensuring accurate context for RAG.
🦾 Namespace-Aware User Message Event: Implemented a new event type (NamespaceAwareUserMessageEvent) that extends the standard user message to include pre-selected knowledge source namespaces, enabling RAG agents to receive and process this crucial contextual information.
📄 Core Namespace Utilities: Developed foundational data structures (BucketNamespacePair) and utility functions (filter_retrievers_by_namespace) within aihub_lib to support dynamic filtering of RAG retrievers based on user-selected namespaces.
⚙️ New Configuration Options: Added NamespaceSelectionAgentConfig and RAGDelegationConfig to provide flexible configuration for the namespace selection workflow, including LLM settings, available buckets, and how it delegates to RAG agents.

Changed

🔄 RAG Agent Namespace Filtering: Updated both the RAG Agent and Expert RAG Agent to leverage the new NamespaceAwareUserMessageEvent. These agents can now dynamically filter their knowledge retrievers based on the pre-selected namespaces, significantly improving the precision and relevance of information retrieval.

Fixed

🐛 OpenAI Service Header Injection: Corrected an issue in the OpenAI service where additional headers were not consistently injected into SDK calls, ensuring proper instrumentation and metadata propagation for requests.

[v0.256.14] - 2026-01-08 - Robust Timeout Management and Enhanced Dagster Run Monitoring

Added

✨ Introduced Comprehensive Document Loading Timeout: Added an OPERATION_TIMEOUT setting for the DoclingLoader, allowing control over the overall duration of document loading operations to prevent indefinite hangs.
🚀 Granular HTTP Timeouts for Docling API: Implemented explicit connect, read, write, and pool timeouts for HTTPX clients used in Docling API interactions, providing more fine-grained control over network requests.
🦾 Explicit Filesystem Timeouts: Configured explicit connection and read timeouts for Azure Data Lake and S3 filesystem resources, improving reliability for storage operations.
⚡️ Enabled Dagster Run Monitoring: Activated Dagster's run monitoring across all environments, with configurable settings for maximum runtime, start timeout, cancel timeout, and poll intervals to automatically manage and prevent stuck runs.

Changed

🔄 Clarified Docling API Timeout Scope: The API_TIMEOUT setting for Docling API calls is now explicitly defined as the timeout for individual API calls, distinguishing it from the new overall operation timeout.

Fixed

🐛 Improved Docling Async Poll Error Handling: Refined the error handling for asynchronous polling in the DoclingLoader to catch more specific HTTP, OS, and asyncio cancellation errors, leading to more robust operation.

Refactor

🧹 Streamlined DoclingLoader Asynchronous Logic: Refactored the asynchronous document loading method in DoclingLoader by extracting core logic into an internal helper method, improving readability and maintainability.

[v0.256.13] - 2026-01-08 - Enhanced Observability for Document Parsing

Changed

📄 Improved Document Parsing Logging: Added comprehensive logging within the parse_document_from_data_lake operation to provide greater visibility and easier debugging for document loading, parsing, and metadata application steps.

[v0.256.12] - 2026-01-07 - Improved Database Connectivity for Docker Environments

Added

🔗 Configured MONGO_CONNECTION_STRING: Explicitly defined the MongoDB connection string across Docker Compose configurations, leveraging ferretdb and existing environment variables (MONGO_USERNAME, MONGO_PASSWORD) for a more robust and clearer database setup.

[v0.256.11] - 2026-01-06 - Enhanced Document Processing and Interactive AI

Added

✨ Introduced Chat-style Human-in-the-Loop (HITL) interactions: Agents can now prompt users with questions directly within the chat interface, appearing as regular messages rather than pop-up dialogs, for more seamless conversational workflows.
🧹 Automatic Docling Result Cleanup: Implemented client-side and server-side mechanisms to automatically clear old Docling conversion results after retrieval, improving storage management and preventing orphaned data.
⚙️ Configurable Docling Result Removal Delay: Added a new CLEAR_RESULTS_DELAY setting to control how long Docling results are retained before automatic cleanup.

Changed

🛡️ Improved Docling Conversion Resiliency: Enhanced error handling for Docling document conversion to more robustly manage network issues, connection resets, timeouts, and server restarts, reducing transient failures.
📊 Enhanced Docling Conversion Logging: Significantly expanded debug and error logging within the Docling loader to provide greater visibility into the document conversion process, aiding in troubleshooting.
🚀 Updated Docling Service Configurations: Modified Docker Compose templates to enable SINGLE_USE_RESULTS and set a RESULT_REMOVAL_DELAY for the Docling service, complementing client-side cleanup efforts.

[v0.256.10] - 2026-01-05 - Improved PDF Handling and Development Workflow

Added

🦾 Enhanced PDF Document Processing: Introduced a new pre-processing step to automatically fix malformed PDF files, specifically those with missing or invalid page dimensions, using A4 as a standard fallback. This significantly improves the robustness of document conversion through the Docling service.
🧪 Comprehensive PDF Pre-processing Tests: Added a dedicated test suite for the new PDF pre-processing functionality, ensuring its reliability and correctness across various PDF formats.
⚡️ Asynchronous Test Utility: Implemented a new run_with_event_loop helper function to streamline the execution of synchronous methods within an asyncio event loop, particularly for Milvus-related testing, enhancing test reliability and developer experience.

Changed

💬 Improved Docling Task Error Reporting: Updated error messages for failed or skipped Docling conversion tasks to provide the full API response, offering more detailed context for debugging and troubleshooting.

Refactor

🔄 Streamlined Dependency Management: Standardized aihub_lib dependency across all projects to use local path references (path = "../aihub_lib", develop = true), simplifying local development and monorepo integration.
🧹 Centralized Asynchronous Test Setup: Refactored Milvus-related test fixtures and helper functions to leverage the new run_with_event_loop utility, centralizing and improving the management of asynchronous operations in tests.

[v0.256.9] - 2026-01-05 - Introducing Chat-Style Human-in-the-Loop for Seamless Agent Collaboration

Added

✨ New Chat-Style Human-in-the-Loop (HITL) Interaction: Agents can now ask questions that appear as regular chat messages, allowing users to respond by typing a normal chat message, enabling more fluid conversational workflows.
🦾 HitlDemoAgent: A new demo agent has been introduced to showcase all three Human-in-the-Loop types: input, confirmation, and the newly introduced chat-style interaction, providing clear examples for developers.
🚀 API Endpoint for Open Chat HITL: A new /threads/{thread_id}/open-chat-hitl API endpoint has been added, allowing client applications to efficiently query for active chat-style HITL requests within a thread.
📄 Dedicated Chat HITL Event Types: New event classes (HumanInTheLoopChatRequestEvent, HumanInTheLoopChatResponseEvent) and a helper class (HumanInTheLoopChat) have been implemented to specifically support and simplify the integration of chat-style Human-in-the-Loop interactions.
🌐 Internationalization Support for Chat HITL: Translations have been added for the new chat-style HITL request and response events across supported languages (DE, EN, FR, IT), ensuring a global user experience.
⚙️ IntelliJ Run Configuration for HitlDemoAgent: A new IntelliJ IDEA run configuration has been included to simplify the setup and execution of the HitlDemoAgent for local development and testing.

Changed

🔄 OpenWebUI Pipeline Intelligent HITL Handling: The OpenWebUI pipeline now intelligently detects active chat-style HITL requests at the start of a user message and routes the user's input as a chat HITL response, streamlining conversational turns.
⚡️ Enhanced Base HITL Events: The core HumanInTheLoopRequestEvent and HumanInTheLoopResponseEvent have been updated to incorporate the new chat interaction type and now utilize type generics for improved type safety and clarity.
🔐 Improved Thread API Access Control: Thread-related API endpoints in the ThreadController now leverage extracted helper methods for more robust, consistent access checking and error handling across operations.
🔌 Application and Playground Router Integration: The new API endpoint for retrieving open chat HITL requests has been seamlessly integrated into the main application and playground router configurations.

Refactor

🧹 Reorganized Human-in-the-Loop Modules: The HumanInTheLoopInput and HumanInTheLoopConfirmation helper classes, along with their respective request/response event classes, have been refactored and moved into dedicated, separate modules for better organization and maintainability.
🏗️ Streamlined HITL Event Imports: Imports for Human-in-the-Loop event types have been simplified and made consistent across the agent codebase and playground examples, reducing verbosity and improving code readability.

[v0.256.8] - 2025-12-30 - Enhanced Document Upload Security and Validation

Fixed

🔒 Strengthened Filename Validation: Implemented more robust and secure filename validation to prevent the upload of malformed or potentially malicious filenames, including those containing path traversal sequences or invalid control characters. This significantly enhances the security and integrity of uploaded documents.
🐛 Improved File Extension Handling: Revised the logic for parsing and validating file extensions, ensuring that all uploaded filenames consistently include a valid, single extension and preventing issues arising from improperly formatted document types.

[v0.256.7] - 2025-12-23 - Enhanced Agent Context Evaluation and Retrieval

Added

✨ Improved Context Management Visibility: The agent now provides clearer insights into its decision-making process, indicating when it has gathered sufficient information to answer a question or when it needs to perform additional retrieval steps to find more relevant data.

[v0.256.6] - 2025-12-19 - Optimized Streaming for Real-time Interactions

Changed

🚀 Improved Streaming Robustness: Enhanced server-sent event (SSE) and streaming responses across chat completions and agent event endpoints by adding critical HTTP headers. This prevents caching and buffering issues from intermediary proxies, ensuring more reliable, real-time data delivery and a smoother user experience.

[v0.256.5] - 2025-12-19 - Enhanced Docling Reliability with Automatic Retries

Added

✨ Docling API Retry Mechanism: Introduced an automatic retry mechanism for Docling API calls, significantly improving resilience against transient network errors and temporary service unresponsiveness during document processing.
⚙️ DOCLING_HTTP_RETRIES Configuration: Added a new configuration setting DOCLING_HTTP_RETRIES (defaulting to 3) to control the number of retries attempted for Docling API requests, providing more control over system behavior.

Changed

🚀 Improved Docling Request Resilience: Docling document conversion processes (both synchronous and asynchronous) now incorporate retry logic with exponential backoff, automatically handling HTTP connection issues, timeouts, and specific transient API errors.
🛡️ Refined Docling Error Handling: Enhanced the classification of Docling API errors to differentiate between transient failures (which trigger retries) and permanent issues, leading to more robust document processing.

Refactor

🧹 Streamlined Docling API Execution: The underlying HTTP request execution logic for Docling API interactions has been refactored into dedicated internal methods, improving code clarity and maintainability.

[v0.256.4] - 2025-12-19 - Enhanced Document Processing: Configurable Figure Descriptions and Robustness

Added

✨ Configurable Figure Description Generation: Introduced a new option to enable or disable the generation of figure descriptions using a vision LLM, providing more control over document processing, with the feature enabled by default.

Fixed

🐛 Improved Figure Description Robustness: Added a safeguard to prevent errors when generating figure descriptions for documents that lack text content, ensuring smoother processing and logging a warning instead.

Changed

⚙️ Updated Quick Start Pipeline: The my_document_pipeline example now explicitly enables both table refinement and figure description generation, showcasing recommended best practices and demonstrating the new feature.

[v0.256.3] - 2025-12-17 - Enhanced Knowledge Management with Shared RAG Pipelines and Dynamic Buckets

Added

✨ Shared RAG Pipeline: Introduced a new shared_rag_pipeline to process and manage shared knowledge, complete with its own Docker image, Dagster configuration, and deployment setup.
📦 Dynamic Knowledge Bucket Configuration: Added new environment variables (AIHUB_CREATE_DEFAULT_BUCKETS, AIHUB_DEFAULT_BUCKET_NAME, AIHUB_SHARED_BUCKET_NAME, AIHUB_DEFAULT_NAMESPACE_NAME, AIHUB_SHARED_NAMESPACE_NAME) to allow flexible naming and conditional creation of knowledge buckets and namespaces.
🌱 Automated Knowledge Bucket Initialization: Implemented API startup logic to automatically create default knowledge buckets and namespaces in MongoDB, ensuring consistent setup.
📄 New Docker Compose Conventions: Added a section to AGENTS.md outlining best practices for Docker Compose templates and environment variable usage.

Changed

⚙️ RAG Agent Knowledge Retrieval: Updated RAGAgent and ExpertRAGAgent to simultaneously retrieve information from both default and shared knowledge buckets and namespaces, enhancing retrieval capabilities.
🔄 Dynamic Pipeline Datalake Naming: Configured existing RAG pipelines (default_rag_pipeline, playground pipeline) to use dynamically defined datalake container names from AIHubSettings, improving configurability.
💬 Refined Condenser Prompt Instructions: Improved the prompt instructions for the standalone question condenser across all supported languages, focusing on better context resolution and preventing the addition of new information. This leads to more accurate and self-contained queries.
🚀 Conditional S3 Bucket Creation: Modified the S3 initialization script (init-buckets.sh) to conditionally create default and shared knowledge buckets based on AIHUB_CREATE_DEFAULT_BUCKETS settings.
💡 Unified RAG Pipelines Development Setup: Replaced the specific "Default RAG Pipeline Dev" run configuration with a consolidated "RAG Pipelines Dev" setup to streamline development for all RAG pipelines.

[v0.256.2] - 2025-12-17 - Platform-wide Observability Boost

Added

📈 Enhanced System Observability: Integrated AihubInstrumentor across all core agent applications (ExpertAskingAgent, ExpertRAGAgent, LLMWrappingAgent, RAGAgent) and the main API. This crucial update enables comprehensive OpenTelemetry instrumentation, significantly improving distributed tracing and telemetry for better system monitoring and debugging capabilities.

[v0.256.1] - 2025-12-16 - Empowering RAG with Human Expertise: Introducing Expert RAG and Asking Agents

Added

🦾 Introduced Expert RAG Agent: A new agent specifically designed for Retrieval-Augmented Generation with integrated human expert escalation capabilities. This agent facilitates consulting human experts when the knowledge base is insufficient.
💬 New Expert Asking Agent: A dedicated agent to manage the end-to-end process of engaging human experts via communication platforms like Microsoft Teams or Slack, capturing their responses, and integrating them into the knowledge base.
⚙️ Expert Escalation Configuration: Added new environment variables (EXPERT_ASKING_CHANNEL_TYPE, TEAMS_CHANNEL_ID, TEAMS_TENANT_ID, TEAMS_BOT_ID, SLACK_CHANNEL_ID, SLACK_SERVICE_URL) to configure the Expert Asking Agent for communication with human experts.
🚀 Deployment Support for Expert Agents: Integrated expert_rag_agent and expert_asking_agent into the CI/CD pipeline and Docker Compose configurations, enabling seamless building and deployment of these new agents.
📄 Expert Escalation Documentation: Updated quick start and agent-specific documentation to guide users through configuring and utilizing the new Expert RAG and Expert Asking Agents.
🛠️ Agent Test Runner Enhancement: Added ensure_dependent_agent_stream method to the AgentTestRunner to better support testing of agent-in-the-loop delegation scenarios.
🔄 Expert Conversation Formatting Utility: Introduced a new utility function (format_expert_conversation) to consistently format expert chat messages for use as context within agents.

Changed

🔄 RAG Agent Redefined: The original RAG Agent has been streamlined to focus solely on standard RAG functionality, removing its previous expert escalation logic and configuration. Expert escalation is now exclusively handled by the new Expert RAG Agent.
📝 Docstring Guidelines Update: Revised AGENTS.md to update docstring best practices, advising against Args: or Returns: sections for conciseness.
🧪 Playground Test Restructuring: Moved expert escalation-related playground examples and tests from the generic RAG Agent to the new, dedicated Expert RAG Agent's test suite for clearer separation of concerns.

Refactor

🧹 Centralized RAG Logic: Extracted common RAG preconditioning and step functions into new shared modules (aihub_agent/aihub_agent/rag/preconditions.py and aihub_agent/aihub_agent/rag/step_functions.py) to promote code reuse and improve modularity across RAG-based agents.
🗂️ Retriever Module Organization: Refactored the aihub_lib/aihub_lib/generative_ai/retrievers module by moving RetrieverConfig and create_retriever into their own dedicated files for better clarity and maintainability.
⚡️ Simplified Retrieval Event Display: Streamlined the retrieve_from_all_sources utility by removing the direct displayer argument, centralizing thought message handling at the agent step level.

[v0.256.0] - 2025-12-16 - Enhanced Dynamic Partition Naming for Clarity

Refactor

🔄 Improved Dynamic Partition Naming: Dynamic partition definitions for various data sources (documents, SharePoint, and local filesystem) now automatically incorporate the datalake_container_name as a prefix. This ensures unique and contextually relevant partition names, enhancing system clarity and preventing potential naming collisions in complex environments.

[v0.255.6] - 2025-12-12 - Smarter Data Lake Cleanup and Milvus Compatibility

Added

🦾 Introduced flexible document parser loader type: The document parser can now be configured with different loader types (e.g., DOCLING) directly via pipeline definitions, enhancing parsing capabilities.
🗑️ Automated figures folder cleanup: When data lake files are deleted, their associated __figures__ folders are now automatically removed, improving data hygiene and ensuring complete data removal.
📄 Centralized Data Lake path building: A new build_path utility method in DataLakeResource provides a consistent way to construct relative paths within the data lake, including any configured directory prefixes.

Changed

⚡️ Enhanced Milvus vector store deletion: The PartitionAwareMilvusVectorStore delete operation now includes a backward compatibility check, ensuring correct fallback to the base class's delete method when manual partitions are not in use, improving reliability.
🔄 Robust S3 directory operations: The S3 Data Lake client now gracefully handles full S3 URIs (e.g., s3://bucket/) for directory existence, listing, and deletion checks, and ensures complete removal of S3 directory marker objects.

Fixed

🐛 Corrected data lake file removal logic: The process for identifying data lake files to remove now accurately considers the configured directory_name prefix, ensuring that URIs are consistently built to prevent incorrect deletion decisions.

Refactor

🧹 Streamlined figure folder naming: The create_figures_folder_name utility has been refactored for cleaner and more direct URI parsing, reducing reliance on os.path functions.

[v0.255.5] - 2025-12-12 - RAG Agent Rejection Message Customization

Added

✨ Configurable context insufficient prompt: Introduced a new context_insufficient_prompt configuration option for the RAGAgent, allowing customization of the leading message when the agent cannot answer due to insufficient context.

Changed

🔄 Dynamic RAG Agent rejection messages: The RAGAgent now dynamically constructs rejection messages based on the new context_insufficient_prompt configuration, providing more flexible and user-friendly responses for few-shot, context insufficient, or expert rejections.
📄 Flexible internationalized guard rejection prompts: Updated the guard.reject translations to support a dynamic {prompt} placeholder, enabling greater customization of rejection messages across all supported languages.

[v0.255.4] - 2025-12-11 - Smarter Standalone Questions: Multimodal Support and Centralized Prompts

Added

🧪 Comprehensive Tests for Question Condensation: Introduced new test cases for the condense_standalone_question utility, ensuring robust functionality for text-only and multimodal inputs, and proper handling of chat history.

Changed

🖼️ Enabled Multimodal Question Condensation: The condense_standalone_question utility now supports multimodal input (e.g., text and images) by accepting a ChatMessage object directly, allowing agents to understand and rephrase questions that refer to visual content.
⚙️ Standardized Standalone Question Prompt: The prompt for generating standalone questions is now centralized and embedded within aihub_lib, removing the configurable condense_question_prompt option from FewShotAgentConfig and RAGAgentConfig to ensure consistent and optimized behavior across all agents.
🚀 Upgraded LLM Interaction for Condensation: The internal mechanism for condensing questions transitioned from simple llm.predict to llm.chat, utilizing system messages to provide more structured instructions to the LLM for improved accuracy and multimodal processing.
💬 Enhanced User Message Handling: Agents now leverage the last_user_message property to retrieve the complete user input, including any multimodal content, ensuring all relevant information is considered during question condensation.
📄 Improved Standalone Question Prompt Logic: Updated the default prompt translations (lib/prompt.*.yml) for standalone question condensation with explicit instructions, including directives for image analysis and replacing pronouns with precise textual descriptions of visible objects.

Removed

🗑️ Deprecated Configurable Condense Prompt: The condense_question_prompt configuration field has been removed from FewShotAgentConfig and RAGAgentConfig, streamlining agent configuration by relying on the new centralized prompt definition.

[v0.255.3] - 2025-12-11 - Enhanced Milvus Performance and Smarter Data Management

Added

✨ Introduced Partition-Aware Deletion for Milvus: Implemented a new deletion mechanism in PartitionAwareMilvusVectorStore that enables more granular and efficient removal of vector nodes associated with a reference document within specific Milvus partitions.
🦾 Enabled Upsert Mode for Milvus Vector Stores: Configured the Milvus vector store factory to use upsert_mode=True, which automatically overwrites existing nodes when new data with matching IDs is added, streamlining data synchronization and preventing duplicates.

Changed

🚀 Optimized Milvus Configuration for Performance and Resource Management: Significantly adjusted Milvus parameters across all configurations (including rocksmq, queryNode, and quotaAndLimits) to improve memory usage, enhance data eviction policies, and optimize query processing, leading to better stability and efficiency.
🧹 Streamlined Milvus Node Addition Workflow: Updated the VectorStoreIOManager to leverage Milvus's upsert_mode, removing the explicit pre-deletion of nodes before adding new ones. This simplifies the data ingestion pipeline and prevents potential memory leaks.
🎯 Enhanced Reference Document Deletion Op: The delete_nodes_for_ref_doc operation now utilizes partition-aware deletion, ensuring that nodes related to a reference document are targeted precisely within their respective Milvus partitions.
⚙️ Upgraded Core Infrastructure Components: Updated Milvus to v2.6.7 and Etcd to v3.5.25, incorporating performance improvements, bug fixes, and new features from these latest versions.

[v0.255.2] - 2025-12-11 - Unlocking Knowledge: Expert Insights and Advanced RAG for Smarter Agents

Added

✨ Expert Insight Persistence: Expert responses and conversations from the ExpertAskingAgent are now stored as valuable 'insights' in MongoDB, enabling agents to learn from and leverage past expert interactions.
🚀 Modular Multi-Source Retrieval System: Introduced a new, flexible system allowing RAG agents to retrieve information from diverse sources, including traditional knowledge bases and newly persisted expert insights, in parallel.
💬 Specialized Human-in-the-Loop (HITL) Events: Enhanced human interaction capabilities with new event types (HumanInTheLoopInput and HumanInTheLoopConfirmation), providing more explicit and structured dialogues for free-form text input or yes/no confirmations.
🌐 Bot-in-the-Loop Path Entity: A new MongoDB path entity has been added to streamline bot-in-the-loop responses, improving the integration and handling of bot channel interactions.
🧪 Comprehensive Agent Test Suites: New test coverage has been added for the ExpertAskingAgent and expanded for the RAGAgent, specifically addressing expert escalation workflows and the new insight retrieval functionality.
🖥️ Robust MongoDB Connection Management: Implemented enhanced connection management for MongoDB within both agent and process runners, ensuring reliable data persistence and retrieval.

Changed

⚙️ RAG Agent Configuration: The RAGAgent's retrieval configuration has been modernized to utilize the new multi-source retrieval system, replacing the singular retrieve_step_config with a flexible list of retrievers.
🆔 Detailed Expert Identification: ExpertAnswerSufficientEvent and ExpertAnswerInsufficientEvent now include the expert_user_id, allowing for more granular tracking and identification of expert contributions.
🤝 Mandatory Bot-in-the-Loop Responder Information: The responder and user_name fields in BotInTheLoopResponseEvent are now required, ensuring complete traceability of human responses from bot channels.
🔄 OpenWebUI Human-in-the-Loop Handler: The OpenWebUI integration has been updated to fully support the new specialized HumanInTheLoopConfirmation dialogs, providing a richer and more intuitive user experience for agent confirmations.

Refactor

🧹 Centralized Retrieval Logic: Core retrieval utilities and configurations, including RetrieveSummariesConfig and helper functions for node retrieval, have been moved to a shared library (aihub_lib), promoting better code organization and reusability across agents.
⚡️ Streamlined Human-in-the-Loop Event Structure: The internal architecture of Human-in-the-Loop events has been refined, abstracting common Request and Response patterns into base classes and introducing dedicated subclasses for specific interaction types.

[v0.255.1] - 2025-12-11 - Enhanced Database Connection Management with PgBouncer

Added

🚀 PgBouncer Integration: Introduced PgBouncer connection pooling for Dagster deployments (all stages except dev) to improve database connection management and prevent connection exhaustion.
✨ PgBouncer Service: Added a dedicated PgBouncer service with robust health checks and configurable pooling parameters to all non-development Docker Compose configurations.
📄 PgBouncer License Information: Included the ISC license details for PgBouncer in the project's license configuration.

Changed

🔄 Dagster Database Connectivity: Updated Dagster's database configuration in build, latest, local, and nightly stages to connect through PgBouncer (hostname pgbouncer, port 6432) instead of directly to PostgreSQL.
⚡️ Service Dependencies: Configured Dagster's webserver and daemon services in non-development environments to explicitly depend on the pgbouncer service, ensuring correct startup order.

Refactor

🧹 Dynamic Dagster Configuration: Refactored the Dagster configuration template (dagster-config.yml.j2) to conditionally configure database connections: direct to PostgreSQL for dev stage, and via PgBouncer for all other stages.

[v0.255.0] - 2025-12-10 - Enhanced Platform Experience and High-Availability Storage

Added

✨ Enhanced User Interface Descriptions: A comprehensive update across all API routes, introducing multi-language support (German, French, Italian) and more descriptive names for key platform components, such as AI Assistants (formerly Agents), Quality Testing (formerly Evaluation), Activity Log (formerly Events), Knowledge Base (formerly Knowledge), AI Models (formerly Models), Chat (formerly Open WebUI), Workflows (formerly Processes), and Platform Overview (formerly Suite).
🔄 New Document Conversion API: Introduced a new /docling/process API endpoint and corresponding schemas (DocumentConversionMetadata, DocumentConversionResponse) to provide advanced document conversion capabilities, allowing direct processing of documents via the API.
🚀 Integrated etcd with SeaweedFS Filer: Added etcd as the metadata backend for SeaweedFS Filer, enabling high-availability deployments and robust metadata storage for the object storage service.
🎨 Improved Image Display in Documentation: Added CSS rules to invert image colors in light mode within the documentation, enhancing readability and visual consistency.
📦 Globally Available NavigationBoxes Component: Made the NavigationBoxes component available globally in the VitePress documentation theme.

Changed

⚙️ Refined OpenAI API Endpoints: Updated OpenAI-compatible API routes to integrate _with_assistants versions for get_model and chat_completion, streamlining interactions with AI assistants.
📝 Updated Image Generation Model Default: Changed the default model for image generation from dall-e-3 to a more generic image-generation string, allowing greater flexibility in model selection.
📚 Comprehensive Documentation Overhaul: Extensive updates to German and English documentation, including terminology clarifications, vision alignment, updated feature descriptions, and rewritten FAQs to reflect the platform's focus on open-source, data sovereignty, and the new SeaweedFS storage.
💾 Alphabetical Sorting of Services in Platform Overview: The platform overview now sorts services alphabetically by name, improving navigability and user experience.
💄 Theme and UI Enhancements: Minor visual adjustments to the documentation, including updating tip box colors from gray to green and refining gradients and font styles for navigation boxes.

Fixed

🐛 Docs Link Handling: Set ignoreDeadLinks: true in VitePress configuration to prevent documentation build failures due to temporary or external dead links.

Security

🔒 Removed Anonymous S3 Read Access: Hardened security for the SeaweedFS S3 policy by removing anonymous read actions, ensuring all S3 access requires proper authentication.

Refactor

🧹 Internal Development Configuration Cleanup: Removed unused <node-interpreter> setting from .idea/runConfigurations/Docs_Dev.xml.
🔄 Build System Configuration for SeaweedFS: Updated generate_compose.py to include new seaweed-filer-config.toml.j2 templates, ensuring proper generation of SeaweedFS configurations across different deployment stages and hardware types.
🧹 User Mock Data Adjustment: Updated the user mock data to reflect the new My Account localization for service names.

[v0.254.19] - 2025-12-10 - Refined Contextual Chat History Management

Changed

🔄 Adjusted message ordering within the limit_chat_history_with_context utility. Context messages are now placed after the limited chat history but before the final user prompt, optimizing the input structure for generative AI models to potentially enhance comprehension and response quality.

[v0.254.18] - 2025-12-09 - Docling Loader Robustness and Development Workflow Enhancements

Fixed

🐛 Improved Docling Content Robustness: Enhanced the Docling document loader to recursively handle and fix malformed null meta fields within document content, ensuring better compatibility and preventing errors when processing data from older Docling server deployments.

Changed

🔄 Expanded PR Branch Naming: Updated CI/CD rules to allow feature branches prefixed with claude/ to pass semantic pull request checks, streamlining specific development workflows.

[v0.254.17] - 2025-12-09 - Enhanced RAG Agent with Expert Escalation and Streamlined Bot Communication

Added

✨ Expert Escalation to RAG Agent: Introduced a new capability for the RAG Agent to escalate questions to human experts when the retrieved context is insufficient, improving answer reliability for complex queries.
🦾 Unified Channel Configuration for Bot-in-the-Loop: Centralized bot communication setup with a new channel_config for the Bot-in-the-Loop agent, enabling seamless integration with both Slack and Microsoft Teams.
💬 Expert Conversation Context Event: Added ExpertAnswerContextEvent to seamlessly integrate expert responses into the RAG Agent's context for generating more informed answers.
🚫 Expert Rejection Event: Introduced ExpertRejectEvent to gracefully handle scenarios where users decline human expert assistance within the workflow.

Changed

🔄 RAG Agent Expert Escalation Flow: The RAG Agent now includes new steps for user consent, invoking the ExpertAskingAgent, and incorporating expert answers, integrating functionality previously handled by the standalone ExpertGroundedAgent.
🤖 Expert Asking Agent Role Refinement: The Expert Asking Agent's responsibilities have been refined to focus solely on posing questions to experts and validating their responses, removing the previous capability to generate and save knowledge snippets to an OpenWebUI knowledge base.
📝 Improved Expert Answer Sufficiency Logic: Refined the prompt and criteria used by the system to determine whether an expert's answer is sufficient, leading to more accurate routing decisions in multi-turn expert conversations.
📊 Enhanced Bot Logging: Added more detailed logging for the Bot-in-the-Loop service to provide better visibility into conversation handling and thread management across different channels.
🖼️ Refined Image Handling in Pipelines: Updated the image processing logic within the AI Hub Pipeline's MessageConverter to directly pass data URLs, improving compatibility with LlamaIndex's image block handling.
⚙️ RAG Agent Configuration Defaults: Adjusted default settings for the RAG Agent, including enabling context sufficiency checks by default and increasing the maximum number of input tokens, leading to improved performance and robustness.

Fixed

🐛 Docker Compose Configuration: Addressed Docker Compose settings by including LOG_LEVEL environment variables for agents and implementing Traefik labels and local port exposure (8001) for the Bot service, enhancing debuggability and accessibility in various deployment stages.

Removed

🗑️ Deprecated Expert Grounded Agent: The ExpertGroundedAgent and all its associated files and events have been removed, as its core functionality has been refactored and integrated into the RAG Agent.
🧹 OpenWebUI Knowledge Snippet Features: Eliminated the ExpertAskingAgent's capabilities related to generating and persisting knowledge snippets to OpenWebUI, streamlining its core function.
📦 Unused aihub_integration Module: Removed references to the aihub_integration module from IDE configuration files.
✂️ Unused stringcase Dependency: Cleaned up the stringcase dependency from pyproject.toml as it was no longer utilized.

[v0.254.16] - 2025-12-05 - Enhanced Data Versioning for Active Partitions

Changed

🔄 Refined Dynamic Partition Versioning: The data versioning operations for Data Lake, Local File System, and SharePoint files now intelligently filter to generate versions only for actively managed dynamic partitions. This improvement ensures that the system tracks versions exclusively for partitions recognized by Dagster, enhancing efficiency and preventing the generation of version keys for stale or non-existent partitions.

[v0.254.15] - 2025-12-04 - Deeper OpenWebUI Integration, Smarter Document Processing, and Refined Model Management

Added

✨ Enhanced OpenWebUI Interactivity: Introduced a suite of new OpenWebUI functions, enabling native streaming for AI-Hub agent responses, interactive displays for event traces and document sources, and direct OpenAI-compatible access to AI-Hub models directly within OpenWebUI.
🚀 LLM-Powered Document Table Refinement: Implemented a new pipeline resource and operation for intelligent table processing during document ingestion, which uses LLMs to automatically detect and split merged tables and identify complex multi-row headers, significantly improving data extraction accuracy.
📦 Dedicated Production Configuration Template: A new .env.prod template now provides a clear and comprehensive guide for configuring production deployments, enhancing setup reliability and security.
⚙️ Streamlined Development Run Configurations: Added new run configurations for simplified local web development and one-click execution of the default RAG pipeline.
🗃️ Automatic Default Knowledge Base Bucket: The S3 storage service now automatically provisions a defaultknowledge bucket, simplifying initial setup for RAG-enabled applications.

Changed

⬆️ Upgraded Core AI Models: Updated LiteLLM configurations to integrate the latest Azure OpenAI models for chat, vision, and audio (e.g., GPT-4o mini, text-embedding-3-large) and Cohere for reranking, boosting overall model performance and broadening AI capabilities.
📈 Optimized Agent Online Status Reporting: The AI-Hub API now determines agent online status based on last_discovered timestamps stored in the database, reducing overhead and improving responsiveness compared to real-time NATS discovery for every request.
🔑 Improved S3 Public File Access: Enhanced S3 file access services to ensure presigned URLs for uploaded documents are consistently generated with public endpoints and s3v4 signatures, facilitating reliable browser access.
📄 Clarified VLM Prompting Guidelines: Updated Vision Language Model (VLM) prompt instructions across all supported languages to strictly guide models to describe only observable visual content, preventing speculative responses based on text context.
🗄️ Standardized MongoDB UUID Handling: All MongoDB connections are now explicitly configured with uuidRepresentation="standard", ensuring consistent and predictable UUID behavior across the platform.
🐳 Improved Docker Compose Build Process: The Docker Compose build run configuration now includes the --build flag by default, ensuring that service images are rebuilt when upstream changes occur.
⏱️ Increased LiteLLM Startup Robustness: Extended the health check start_period for the LiteLLM service to 120 seconds, improving stability during initial deployment and preventing premature health check failures.
🛠️ Refined RAG Agent and Pipeline Defaults: Adjusted default RAG agent and pipeline configurations, including upgrading to embedding/large and adding max_partitions to improve data ingestion control and vector retrieval efficiency.
🖥️ OpenWebUI Feature Cache Management: Disabled the base model cache in OpenWebUI to ensure that dynamic pipeline models and agent configurations are always fetched fresh.
🐛 Docling Document Loader Table Output: Modified the Docling document loader to output tables in a standardized Markdown format instead of raw HTML, improving compatibility with subsequent document parsing steps.

Refactor

🧹 Centralized Logging Infrastructure: The core logging system was refactored and moved to aihub_lib.infrastructure.logging, promoting a more modular and reusable architectural design.
⚙️ Consolidated Internal Service Endpoints: Internal Docker network endpoints for various services are now hardcoded directly within Docker Compose templates, simplifying environment variable management and improving deployment consistency.
🗑️ Modernized OpenWebUI Integration Module: The aihub_integration module has been removed, as its functionalities are now provided through the new, more flexible OpenWebUI function integration.

Fixed

🐛 Docling Metadata Handling: Resolved an issue in the Docling document loader where null meta fields in JSON content could cause processing errors, enhancing document ingestion reliability.
🐞 WebSocket Endpoint Correction: Fixed an incorrect WebSocket endpoint path in the API service configuration, ensuring proper real-time event streaming for clients.

[v0.254.14] - 2025-12-03 - LLM Input Processing and Multimodal Enhancements

Added

✨ Enhanced UserMessageEvent: Introduced the last_user_message property to UserMessageEvent, providing direct access to the complete last user message, including all multimodal blocks (like images), for richer LLM interactions.

Changed

🔄 Improved LLM Message Handling in RAG Agent: The RAG Agent now explicitly merges consecutive chat messages (including multimodal content blocks) with the same role right before sending them to the LLM. This ensures better compatibility with various LLM providers (e.g., LiteLLM) and more robust multimodal support.
🦾 Refined Multimodal Message Merging: The internal merge_consecutive_messages utility has been updated to correctly handle and combine multimodal ChatMessage blocks, ensuring proper formatting when multiple parts of a message are from the same role.
🖼️ Simplified Image URL Processing: The Open WebUI integration now streamlines the handling of image URLs, including data:image URLs, by directly passing them to LlamaIndex's ImageBlock. This improves consistency and reliability for multimodal inputs.

Refactor

🧹 Streamlined LLM Configuration: The custom PreprocessingOpenAILike wrapper has been removed, and LLMConfig now directly utilizes OpenAILike. This change centralizes message preprocessing logic within the agent's flow rather than the LLM wrapper, simplifying the overall architecture.

[v0.254.13] - 2025-12-03 - Robust Evaluation and Phoenix Client Modernization

Refactor

🔄 Upgraded Phoenix Client Integration: Modernized the backend to utilize the new asynchronous Phoenix client, streamlining experiment and dataset interactions with updated imports, client initialization, and method calls.
🦾 Optimized LLM Structured Output: Improved the judge LLM's ability to generate structured evaluations by transitioning from direct JSON output instructions to leveraging LlamaIndex's advanced tool-calling capabilities, enhancing reliability.

Changed

⚡️ Stricter Evaluation Summary Generation: Evaluation summaries are now only created when valid scores are available, ensuring data integrity and preventing empty summary entries in the evaluation results.
📄 Updated Evaluation Summary Data Model: The avg_score field in evaluation summaries is now explicitly a required float, aligning with the refined backend logic that guarantees a valid score when a summary exists.

Fixed

🐛 Improved UI Robustness for Missing Scores: Frontend evaluation result displays now gracefully handle cases where scores might be absent for certain evaluators, preventing UI errors and defaulting to zero when data is unavailable.
🛠️ Corrected Phoenix Experiment ID Access: Addressed an issue where RanExperiment data was accessed incorrectly, ensuring reliable retrieval of experiment IDs after an experiment run.

[v0.254.12] - 2025-12-03 - Enhanced Document Processing and Model Updates

Added

✨ Introduced Docling Standard Pipeline: Added support for a new "Standard" Docling document processing pipeline, providing advanced control over OCR, table extraction, image handling, and more flexible output formats.
⚙️ Configurable Docling Asynchronous Polling: New environment variables (DOCLING_POLL_INTERVAL, DOCLING_MAX_POLLS) have been added to allow customization of the polling interval and maximum attempts for asynchronous Docling document processing tasks.

Changed

🚀 Upgraded Default RAG Embedding Model: The default embedding model used by the RAG Agent has been updated from embedding/small to embedding/large to potentially improve retrieval quality and performance.
🔄 Updated Azure OpenAI Model Reference: LiteLLM configurations for text-generation/mini have been adjusted to use azure/gpt-4o-mini instead of azure/gpt-5-mini, aligning with recent Azure OpenAI API changes for model ingestion.
branding Updated OpenAI Compatibility Interface: The AI Hub's OpenAI-compatible interface and web UI are now consistently branded as "Open WebUI," featuring updated icons for better recognition.

Fixed

🐛 Improved Document Parsing Robustness: The Docling document loader now uses html5lib for parsing, enhancing its ability to handle malformed or complex HTML content and improving overall parsing reliability.

Refactor

🧹 Standardized Docling API Configuration: The DOCLING_API_ENDPOINT environment variable has been renamed to DOCLING_BASE_API_URL across the codebase, improving clarity and consistency in Docling service configuration.

[v0.254.11] - 2025-12-02 - Enhanced Document Intelligence Figure Extraction

Changed

🖼️ Document Intelligence Figure Handling: Reworked the figure extraction and replacement mechanism within the DocumentIntelligenceLoader to leverage precise span offsets provided by the API, ensuring more accurate and robust figure integration into the document text.
⚡️ Improved Figure Processing Robustness: Implemented a new strategy for figure replacement that processes figures in reverse order of their appearance, mitigating issues related to text shifts and ensuring correct figure placement.
🐛 Strengthened Figure Data Validation: Introduced validation checks for figure span information to gracefully handle cases where span data is missing or out of bounds, preventing errors during document processing.

[v0.254.10] - 2025-12-02 - AI-Powered Code Review and On-Demand Assistance with Claude

Added

✨ Automated AI Code Reviews: Introduced a new GitHub Actions workflow to automatically run Claude AI code reviews on pull requests, providing comprehensive feedback on code quality, potential bugs, performance, security, and test coverage.
🤖 On-Demand AI Assistant: Added a new GitHub Actions workflow enabling direct interaction with Claude AI within issues, pull request comments, and reviews by tagging @claude. This facilitates on-demand code assistance, explanations, and insights.

[v0.254.9] - 2025-12-02 - Unified File System Filtering with Flexible Patterns

Refactor

🧹 Unified Local File System Scanning: Significant overhaul of LocalFileSystemResource to replace granular folder, subfolder, extension, and file-specific filters with a more flexible, unified include_patterns and exclude_patterns system, simplifying file discovery and filtering logic.
🔄 Generic Source File Types: Deprecated LocalFile and MinimalLocalFile types, transitioning to the more generic SourceFile and MinimalSourceFile types for local file system interactions. This removes file system-specific folder attributes (source_folder, subfolder) for broader applicability.
⚡️ Improved Pattern Utility Functions: Rewrote and enhanced pattern utility functions to support the new flexible include_patterns and exclude_patterns system, providing more robust and explicit ways to generate regex patterns for paths, folders, extensions, and substrings.

Changed

📄 Updated IO Manager and Ops: The LocalFileSystemIOManager and related data versioning operations have been updated to align with the new SourceFile and MinimalSourceFile types and the simplified file system resource configuration.
⚙️ Simplified Local-to-Datalake Definitions: The default_local_filesystem_to_datalake_definitions utility now uses the streamlined include_patterns and exclude_patterns for configuring local file system scanning, making setup more intuitive.
📊 Adjusted Metadata Tables: Local file metadata tables (local_file_metadata_table) no longer display source_folder and subfolder information, reflecting the removal of these attributes from the underlying file objects.

Removed

🗑️ Deprecated LocalFile Types: The specific LocalFile and MinimalLocalFile Python types have been removed in favor of the more generic SourceFile and MinimalSourceFile types.
🗑️ Legacy File System Scan Configuration: Eliminated various specific configuration parameters (e.g., include_folders, exclude_paths, include_extensions) from LocalFileSystemResource and related utility functions, replaced by the unified pattern-based approach.

[v0.254.8] - 2025-12-01 - Smarter Data Fetching for Enhanced UI Stability

Added

✨ Introduced useRouteReady Composable: A new utility function to ensure that data fetching operations are only initiated once all required route parameters are fully resolved and valid, preventing queries from executing with incomplete or placeholder values during navigation.

Refactor

🧹 Optimized Data Query Activation: Implemented the useRouteReady composable across various data fetching hooks, including those for Agents, Documents, Evaluation Datasets and Experiments, Processes, Roles, Threads, and Users. This change prevents API queries from running prematurely, leading to more stable UI states and improved application reliability during route transitions.

[v0.254.7] - 2025-11-28 - New Service Integration: Attu Platform

Added

✨ Attu Service Integration: Introduced the Attu service to the platform, enhancing capabilities (e.g., for data management).
🔑 Secure Access for Attu: Implemented OAuth2 Proxy for the Attu service, ensuring secure, group-based access for authorized users.
🚀 Automated Routing for Attu: Configured Traefik to provide automatic routing and SSL termination for the Attu service via a dedicated subdomain.

Changed

📄 Updated OAuth Configuration Documentation: The quick start guide has been updated to include the required redirect URIs and platform type configurations for the new Attu service.
⚙️ Deployment Configurations for Attu: Modified various Docker Compose files and templates to incorporate the Attu service and its OAuth2 Proxy into the deployment stack.

[v0.254.6] - 2025-11-24 - Documentation Experience Refresh and Strategic Messaging Update

Added

🖼️ Enhanced Image Viewing: Integrated a new lightbox plugin and medium-zoom functionality into the documentation, providing an improved and interactive experience when viewing images.
⚡️ Custom Documentation Layout: Introduced a dedicated Layout.vue component to streamline and centralize documentation theme customizations, including a GradientBackground component.

Changed

📄 Updated Homepage Features: The homepage (index.en.md) feature descriptions have been entirely rewritten to better articulate the platform's value proposition, emphasizing open-source, ownership, rapid deployment, strategic independence, transparency, and collaborative strength.
📐 Refreshed Architecture Diagrams: Updated various high-level and low-level architecture diagrams to reflect the latest system design and clarity.

Refactor

🧹 Documentation Theme Structure: Refactored the documentation theme's index.js to leverage a dedicated Layout.vue component, simplifying the theme configuration and improving maintainability.
🔄 Dependency Import Path: Adjusted the import path for CopyOrDownloadAsMarkdownButtons in the documentation theme for consistency and clarity.

Changed

📄 Updated DNS Prerequisites: Replaced the datalake-api subdomain requirement with litellm.aihub.example.com for the litellm proxy, streamlining API access configurations for the platform.

Removed

🗑️ Removed Docling Model Initialization Script: The dedicated init-models.sh script, previously responsible for checking and downloading Docling models, has been removed, indicating an updated or integrated approach to model management.

[v0.254.4] - 2025-11-21 - Internal Consistency Improvements

Refactor

🧹 Standardized build_uri parameter: Updated the data_lake_client.build_uri method call within the data lake file removal process to use file_path instead of path, enhancing parameter naming consistency across the API.

[v0.254.3] - 2025-11-18 - Unlocking Multi-Tenancy and Streamlined Onboarding with Enhanced Platform Clarity

Added

✨ Introduced comprehensive Multi-Tenancy documentation: New sections detail the multi-tenancy concept, how to set up tenants, manage users and roles, and a technical reference for access control, providing clear guidance on organizational boundaries within a single platform instance.
🚀 New Quick Start and One-Command Deployment guides: Detailed documentation now covers prerequisites for both production and local deployments, Azure Entra ID setup, and a streamlined one-command deployment process for rapid platform setup.
📄 Expanded manual Slack & Teams Bot Creation guide: A new, in-depth guide is available for manually configuring bots, including Teams Developer Portal setup, MongoDB configuration, Slack API integration, app manifest examples, and troubleshooting.
⚙️ Detailed RAG Ingestion Pipeline documentation: New content explains the processing stages (parsing, chunking, embedding, linking, summarization), storage, document lifecycle, and benefits of the RAG ingestion pipeline.

Changed

🔄 Refined Language Model (LLM) integration documentation: Updated content clarifies LLM proxy functions, model configuration, PII protection, and guardrail concepts, enhancing understanding of AI model interactions.
📈 Improved Cost Control documentation: Enhanced details on LiteLLM's user management for budgets and rate limiting, including configuration via environment variables, to provide better predictability and control over AI expenses.
📚 Clarified deployment terminology: Documentation for deployment options, backup & recovery, and updates & maintenance now explicitly differentiates between "Multi-Instancing" (hard infrastructure separation) and "Multi-Tenancy" (logical separation within an instance).
🦾 Updated Agent documentation: The fundamentals, RAG Agent, and Expert Asking Agent sections have been extensively rewritten to provide deeper insights into structured workflows, context management, and human-in-the-loop collaboration patterns.
💡 Enhanced Data Pipeline and Knowledge Management documentation: Rewritten sections offer improved clarity on pipeline fundamentals, workflow automation, and knowledge organization through "collections" (namespaces), detailing their structure, lifecycle, and integration.
🌐 Updated main page and FAQ: The main landing page and Frequently Asked Questions have been revised to reflect the latest platform features, documentation structure, and terminology.
🧹 Standardized documentation terminology: A new glossary has been added to the translation prompt to ensure consistent terminology usage across all localized documentation, especially for multi-tenancy and related concepts.

Removed

🗑️ Node.js dependencies: The platform no longer includes Node.js packages, simplifying the dependency stack.

Refactor

📁 Restructured documentation directories: Several top-level documentation directories (e.g., 15_slack_teams_integrations to 16_slack_teams_integrations, 16_api to 17_api) have been renamed to accommodate the new Multi-Tenancy section and improve logical grouping.
🔗 Updated internal documentation links: All internal links within the documentation have been updated to reflect the new directory structure and ensure navigation consistency.
📦 VitePress plugin import simplification: Updated the import path for CopyOrDownloadAsMarkdownButtons in the documentation theme for better maintainability.

[v0.254.2] - 2025-11-18 - New Service Integrations and Deployment Enhancements

Added

🚀 Gunicorn Dependency: Incorporated Gunicorn, a robust WSGI HTTP server, into the project dependencies to bolster production deployment capabilities.
📄 Docling Configuration: Integrated new environment variables for Docling and Hosted VLM API endpoints, timeouts, and model specifics across all Docker Compose configurations, preparing the ground for advanced document processing and visual language model integrations.

[v0.254.1] - 2025-11-18 - Milvus Partitioning for Memory Efficiency and System Enhancements

Added

✨ Milvus Manual Partitioning: Introduced a new manual partitioning system for Milvus vector stores, enabling namespace-based data isolation. This allows for more targeted and memory-efficient loading of data during queries by only accessing relevant partitions.
⚙️ PartitionAwareMilvusVectorStore: A new Milvus vector store implementation that transparently routes data to specific, hashed partitions based on namespace metadata during insertion and queries, significantly improving memory management for RAG workloads.
🧪 Comprehensive Milvus Partitioning Tests: Added a new suite of behavioral tests to validate the correctness and efficiency of the Milvus manual partitioning and namespace routing logic.
🔑 Azure OpenAI Base URL Configuration: Added AZURE_OPENAI_BASE_URL to development environment configurations, providing greater flexibility for connecting to various Azure OpenAI service endpoints.
🔒 Enhanced OAuth2 Proxy Scopes: Expanded the requested scopes for OAuth2 Proxy to include openid, email, and profile, allowing for more comprehensive user information retrieval during authentication.

Changed

🚀 Optimized Milvus Memory Mapping (mmap): Updated Milvus query node configurations to broadly enable memory mapping for vector indexes, scalar fields, and growing data segments. This change improves disk I/O efficiency and reduces RAM footprint, with a potential slight increase in query latency for data not in memory cache.
🔄 Milvus Vector Store Factory Streamlining: The create_milvus_vector_store factory and its associated Dagster resources no longer require manual num_partitions or enable_mmap parameters, as partitioning is now automatically handled, simplifying vector store creation.
📦 Dagster Pipeline Dockerfile Refinement: Streamlined the Dagster pipeline Dockerfile by removing the make build tool dependency and explicitly defining entrypoints in Docker Compose, enhancing deployment clarity.

Refactor

🧹 Dagster Summary Nodes Asset Path: Reorganized the asset key for summary nodes in Dagster pipelines, moving them into a more logical datalake_to_vectorstore subgroup for improved asset discoverability and organization.

Removed

🗑️ Redundant Dagster Volume Mounts: Eliminated redundant dagster-data volume mounts from several Dagster-related services in Docker Compose configurations, reducing clutter and simplifying deployment setups.

[v0.254.0] - 2025-11-18 - Enhanced Data Persistence and Pipeline Management

Added

✨ Persistent Data Storage for Dagster: Implemented persistent volume mounts (dagster-data) for Dagster services, including dagster-webserver, dagster-daemon, and default-rag-pipeline. This change ensures that critical Dagster operational data, such as run history and event logs, is preserved across container restarts and updates.
🛠️ make Utility in Dagster Pipelines: The make utility is now installed within the aihub_pipeline Docker image, providing enhanced capabilities for scripting and managing pipeline operations.

Changed

🔄 Streamlined Dagster Webserver Entrypoint: The aihub_pipeline Dockerfile now explicitly defines make dagster-webserver as its entrypoint, simplifying and standardizing the launch process for the Dagster webserver within the container.
🔒 Simplified OAuth2 Proxy Configuration: The explicit OAUTH2_PROXY_SCOPE setting has been removed from the OAuth2 proxy configuration, streamlining its setup while maintaining robust authentication flows.

[v0.253.4] - 2025-11-18 - Next-Gen Process Management: Introducing Walkthroughs and Advanced Form Capabilities

Added

🚀 Detailed Process Walkthroughs API: Introduced a new endpoint (/processes/{process_class}/{process_id}/walkthroughs) to retrieve paginated process walkthroughs with comprehensive step information, offering deep visibility into process execution.
✨ Process Walkthrough Logic: Implemented robust backend logic within ProcessService to aggregate and construct detailed ProcessWalkthroughDTO and ProcessStepDTO objects, including involved agents and humans, from raw event data.
📦 New Process Input Specification DTOs: Created specialized Data Transfer Objects (AgentInDTO, HumanInDTO, ProgramInDTO) for process input specifications, enhancing structure and type safety for API definitions.
✨ Automatic Formkit Field Configuration: Enhanced the Form class to automatically assign id and required properties to Formkit elements based on Pydantic field definitions, simplifying form creation.
🚀 Extended Formkit Element Library: Introduced a comprehensive suite of new Formkit elements (e.g., DatePicker, MultiSelect, Slider, Password, InputMask, Knob, Rating, Textarea, Checkbox, RadioButton, Select, SelectButton, ToggleSwitch, InputNumber, InputOtp, CascadeSelect), significantly expanding UI capabilities for user input.
✨ Added ref to Formkit Elements: Introduced a ref field (aliased as id) to FormkitElement, providing a unique identifier for better UI control and integration.
🔗 Event Response Tracking: Added an in_response_to field to WorkEvent and Human form submissions, improving the traceability and context of work events within process walkthroughs.
👤 Agent Submission Tracking: Introduced a submitted_by field in AgentWorkEvent, allowing clear identification of the agent responsible for submitting work.
📊 Paginated Process Walkthrough Retrieval: Implemented get_paginated_walkthrough_events to efficiently retrieve process walkthroughs with all their associated events, supporting pagination for large datasets.
🆕 Process Card UI Component: Introduced a new Vue component for displaying process information cards, enhancing the visual representation of available processes.
📝 Reusable Formkit Form Component: Developed a versatile Vue component (Process/Form.vue) for rendering dynamic Formkit schemas, complete with conditional field logic and label interpolation, enabling flexible user input.
🚀 Process Start Forms Component: Created a new UI component (Process/Starts.vue) to render and manage multiple start forms for a process, simplifying the initiation of new process runs.
📊 Process Walkthrough List Component: Implemented a new UI component (Process/Walkthrough/List.vue) to display paginated process walkthroughs, featuring details like creation/update times, involved entities, and step progress visualization.
💻 useProcessWalkthroughs Composable: Introduced a new Nuxt composable (useProcessWalkthroughs) for fetching and managing paginated process walkthrough data, simplifying data access in the frontend.
🚀 useSendProcessStartForm Composable: Created a new Nuxt composable (useSendProcessStartForm) to handle the submission of process start forms, integrating with the query cache for immediate UI updates.
🌐 New Process Walkthrough Translations: Added new English translations for process management and walkthrough-related UI elements.
🛠️ Formkit PrimeVue Integration: Integrated the formkit-primevue module and defined a comprehensive list of wrapped PrimeVue inputs, preparing for advanced form rendering capabilities.
🚀 New Process Detail Pages: Introduced a new set of pages for process details, including an overview, a dedicated "start run" form, and a list of walkthroughs, enhancing process management capabilities.
✨ New SubmittedCV Human Start Event: Introduced a new SubmittedCV event, acting as a human-initiated process start event with a rich set of Formkit input fields for comprehensive CV submission.
🚀 Added Agentic CV Process Runner: Introduced a dedicated runner (run.py) for the AgenticCVProcess playground example, facilitating easier local execution and testing of the workflow.

Changed

⚙️ Docker Compose Dev Configuration: Updated the Docker Compose Dev run configuration to explicitly include .env file paths, ensuring environment variables are correctly loaded.
🔄 Refined Process API Responses: Applied response_model_exclude_none=True to several process-related endpoints, streamlining API responses by omitting null values.
📄 Renamed Process DTOs: Renamed ProcessHumanInDto to HumanInDTO for consistency and clarity in process input definitions.
⚙️ Simplified Human Input Retrieval: Refactored get_process_start_forms and get_process_open_forms to directly return HumanInDTO objects, simplifying the API and reducing data transformation.
📄 Localized Process Configuration from Entity: Added a new constructor to ProcessConfigDTO to facilitate creating localized process configurations directly from entity specifications.
🔄 Standardized Process Input DTOs: Migrated ProcessDTO to use new, specialized DTOs for human, program, and agent inputs (HumanInDTO, ProgramInDTO, AgentInDTO), enhancing type safety and clarity.
🔄 Updated Formkit Element Usage: Migrated InputTextElement to the new InputText class across various event definitions and process runners, aligning with the formkit refactor.
⚙️ Refined Agent Work Event Creation: Updated agent work event creation to include submitted_by information, enhancing event context.
📄 Enhanced Process Entity Specifications: Upgraded process entity specifications (ProgramInSpecsEntity, HumanInSpecsEntity, AgentInSpecsEntity) to accept type-hinted specs objects directly, improving type safety.
🌐 Localized Process Configuration Persistence: Updated ProcessConfig entity to persist name and description as LocaleStringEntity, enabling robust internationalization for process configurations.
🎨 Updated Default Process Icon: Changed the default icon for new processes to carbon:ibm-event-processing, providing a more relevant visual representation.
🏗️ Simplified Dockerfile for Pipelines: Streamlined the Dockerfile for pipelines by removing unnecessary make dependency and delegating entrypoint definition to Docker Compose, improving container efficiency.
🔗 Improved Agent Delegator Event Tracking: Enhanced AgentDelegator to include in_response_to and submitted_by in agent work events, and aligned display_id with the event's unique ID for consistent event tracing.
🤖 Agentic CV Process Overhaul: Significantly refactored the AgenticCVProcess playground example to utilize an expanded range of Formkit elements for initial human input, enhancing the interactive experience.
🔄 Refined Agentic CV Process Logic: Updated the AgenticCVProcess to align with new event tracking mechanisms and improved agent output parsing, ensuring smoother workflow execution.
⚙️ Refined AnalyzedCV Event Structure: Updated AnalyzedCV event to specifically use LLMStopEvent for agent work responses and removed the cv_name field for a more generic approach.
⚙️ Restricted Human Output in Agent-to-Human Process: Updated AgentToHumanProcess and HumanOnlyProcess to target the AIHubAdmin role for human output, enhancing control over human-in-the-loop interactions.
🖥️ Updated Processes Overview: Integrated the new ProcessCard component into the processes overview page, providing a visually enhanced list of available processes.
⚙️ Enhanced PrimeVue Element Validation and Labeling: Improved PrimeVueElement to automatically include required validation rules and append an asterisk (*) to the labels of required fields when localized.
⚙️ Enhanced useTimeAgo Flexibility: Updated the useTimeAgo composable to accept numeric timestamps (nanoseconds) in addition to strings and Date objects, increasing its versatility.
⚙️ Optimized PrimeVue Auto-Import: Configured PrimeVue to exclude wrappedPrimeInputs from auto-import, ensuring these components are managed through Formkit.
🛠️ Extended API Endpoints for Process Walkthroughs: Generated new API endpoints and types (GetProcessWalkthroughsData, GetProcessWalkthroughsResponse) to support fetching paginated process walkthrough data.
📦 Comprehensive Formkit Element Schemas: Added schemas for an extensive range of new Formkit elements, providing full API support for rich user input forms.
📄 Source Origin Tracking for Ingested Nodes: Introduced a source_origin field to IngestedNode, allowing better traceability of original document sources in data lake operations.
🔄 Standardized Process Input Schemas: Updated API schemas for process inputs to align with the new AgentInDTO, HumanInDTO, and ProgramInDTO structures.
🎙️ Expanded Audio Voice Options: Added new voice options (marin, cedar) to ChatCompletionAudioParam, offering more choices for audio generation.
🔑 Enhanced OAuth2 Proxy Scopes: Added openid email profile scopes to the oauth2-proxy configuration, improving user identity and profile information retrieval for Dagster.

Fixed

🐛 Temporarily Disabled Failing Test: Commented out test_walk_through_process_std_methods in the API playground as it currently fails due to flawed logic, to be re-enabled after fix.
🧪 Marked Flaky Test: Added @pytest.mark.flaky to test_stream_response in test_ChatBot.py to identify and manage intermittent test failures.

Refactor

🧹 Import Reorganization: Standardized import order across DoclingController, DoclingService, EventController, and Azure Data Lake resources for improved code consistency and readability.
🧹 Import Normalization: Converted relative imports to absolute imports within OpenaiController for improved modularity and clarity.
🧹 Pydantic Model Validation Update: Migrated event deserialization to use .model_validate() method for Pydantic models, enhancing strictness and correctness in data parsing.
🧹 Simplified Logging in Process Runner: Streamlined logging messages in ProcessRunner by removing redundant process ID information, making logs cleaner.
🧹 Asynchronous Process Steps: Converted received_cv_2_analyzed_cv to an asynchronous method, improving performance for I/O-bound operations.
🧹 Documentation Formatting: Applied minor formatting adjustments to the bot creation guide for improved readability.
🧹 Standardized Agent Route Structure: Refactored agent-related routes to use a more consistent /[agent_class]-[agent_id] format, improving URL readability and maintainability.

Removed

🗑️ Removed Redundant Human Input DTO: Eliminated ProcessHumanInDto as its functionality is now covered by the more granular HumanInDTO.
🗑️ Removed Program-Initiated SubmittedCV Event: Deprecated and removed the program-initiated SubmittedCV event, transitioning the AgenticCVProcess to primarily use human-initiated start events.
🗑️ Streamlined Dagster Volume Mounts: Removed explicit volume mounts for Dagster data from docker-compose configurations, simplifying deployment and relying on other means for data persistence if needed.

Changed

⚙️ Adjusted SeaweedFS Volume Size Limit: The default volume size limit for SeaweedFS has been reduced to 29GB (from 1TB) for non-development environments, optimizing storage resource allocation.
🔄 Externalized API and OTel Version Configuration: The AIHUB_API_VERSION and OTEL_RESOURCE_SERVICE_VERSION are now configurable via environment variables, providing greater flexibility for deployment and consistent version management across services.

Removed

🗑️ Disabled Automatic Speaches Model Downloads: The speaches service no longer automatically downloads models on container startup. This change requires models to be provisioned manually or via an alternative mechanism.
🗑️ Deprecated Dagster Pipeline Configuration: The default_rag_pipeline configuration for Dagster has been removed, streamlining the Dagster workspace setup.

[v0.253.2] - 2025-11-14 - Comprehensive Bot Setup Documentation

Added

📄 New Bot Creation Manual Setup Guide: Introduced a comprehensive, step-by-step guide for manually creating and configuring bots with Microsoft Teams and Slack. This detailed resource covers everything from Azure Bot Framework and Teams Developer Portal setup to MongoDB configuration and Slack OAuth integration, complete with app manifest examples, key terminology, and troubleshooting tips.

Changed

🔗 Updated Chat Interface Documentation Link: Corrected and updated an internal documentation link pointing to the chat interface feature overview, ensuring users are directed to the most current information.

[v0.253.1] - 2025-11-14 - Robust Milvus Vector Store Configuration and CI/CD Streamlining

Added

✨ Introduced Advanced Milvus Vector Store Configuration: The Milvus vector store factory now supports configurable index types (HNSW, DISKANN, IVF_FLAT, FLAT), enabling tailored performance and memory usage for different RAG workloads.
🚀 Enabled Hybrid Search with BM25: Milvus collections now support hybrid search by automatically creating sparse vector indexes (BM25) alongside dense embedding indexes, significantly improving retrieval relevance.
🔑 Implemented Milvus Partition Keys: Vector stores are now partitioned by namespace, enhancing data isolation and query performance for multi-tenant or multi-document RAG applications.
⚙️ Added Memory Mapping (mmap) Option for HNSW: Users can now enable mmap for HNSW indexes in Milvus to reduce RAM consumption, offloading data to the OS page cache for memory-constrained environments.
📄 Added regenerate_compose CI/CD Input: A new input allows explicit control over Docker Compose file regeneration in CI/CD workflows, optimizing build times when regeneration is not needed.

Changed

⚙️ Exposed Milvus Configuration in Dagster Resources: Milvus vector store parameters like index type, mmap, and partitions are now exposed as configurable options in Dagster resources, allowing pipeline authors to fine-tune their RAG infrastructure.
🧪 Improved Milvus Test Isolation: Milvus test fixtures now ensure a clean state by dropping collections and clearing the Milvus vector store factory cache before each test run, enhancing test reliability.
🔄 Optimized CI/CD Dependency Installation: The order of global dependency installation in CI/CD has been adjusted to precede Docker Compose file generation, ensuring all prerequisites are met for a more robust build process.

Refactor

🧹 Refined Milvus Type Hints: Updated internal type hints for Milvus vector store configurations, improving codebase clarity and maintainability.

[v0.253.0] - 2025-11-14 - Unified Deployment, Local AI Power, and Docling VLM Integration

Added

✨ Dynamic Deployment System: Introduced a new Jinja2-based templating system to generate Docker Compose and service configuration files dynamically across various stages (dev, local, build, nightly, latest) and hardware (CPU, GPU), replacing manual configurations for enhanced flexibility and maintainability.
🦾 Local Inference Services: Integrated new Docker services (llama-cpp, llama-cpp-embedding, llama-cpp-reranker, vllm-docling, speaches) to enable on-premise inference for Large Language Models, embedding models, reranker models, and audio models (SST/TTS), dynamically selected based on hardware and deployment stage.
🖼️ Docling VLM Integration: Implemented new API endpoints and a dedicated controller within aihub_api to support advanced document conversion to Markdown using Docling's Visual Language Model (VLM) capabilities.
⚙️ IntelliJ Run Configurations: Added new IntelliJ IDEA run configurations for Docker Compose (Build, Dev, Local with CPU/GPU options) to streamline local development setup and execution.
📄 Deployment Documentation: Created a new README.md within the deployment directory, providing comprehensive documentation for the different deployment configurations and Traefik setup.
🔑 Hugging Face API Key Management: Integrated Hugging Face API key support into CI workflows and LiteLLM configurations, enabling seamless access to models requiring authentication from Hugging Face Hub.

Changed

🛠️ Centralized Configuration: Refactored core infrastructure service configurations (LiteLLM, Milvus, NATS, Dagster, OpenTelemetry, Traefik) into a centralized compose-config.yml for consistent and dynamic generation.
🚀 LiteLLM Model Definitions: Updated LiteLLM proxy to dynamically route to local inference services (llama-cpp, speaches) based on hardware availability and deployment stage, optimizing resource utilization.
🛡️ LiteLLM Prompt Injection Detection: Enhanced LiteLLM's prompt injection detection by disabling simple heuristic checks and enabling more robust similarity and LLM-based API checks with a refined system prompt.
📊 OpenTelemetry Configuration: Implemented conditional activation of OpenTelemetry exporters and processors, ensuring full observability pipelines are enabled only in non-development stages for optimized resource usage.
📦 Milvus Configuration Tuning: Adjusted Milvus logging levels, segment sizes, GPU memory pool settings, and default data handling policies for improved performance and consistency across environments.
⚡️ NATS Performance Tuning: Dynamically configured NATS server settings (payload limits, connections, JetStream storage) to align with specific stage requirements, optimizing performance for both development and production.
🌐 Traefik Configuration: Enhanced Traefik setup for dynamic TLS certificate resolution (Let's Encrypt for production, self-signed for local/build) and dashboard access based on the deployment stage.
📜 Docling Loader Settings: Simplified the DoclingLoader configuration by removing numerous granular settings, streamlining its integration into the document processing pipeline.
🧑‍💻 OpenWebUI Image Generation Model: Changed the default image generation model in OpenWebUI to a generic "image-generation" identifier for broader compatibility with various image generation services.

Fixed

🐛 Agent Test Reliability: Improved the AgentTestRunner by introducing a StopEvent mechanism, allowing tests to complete more rapidly and reliably without relying solely on fixed delays.
🧪 Removed Flaky Test Mark: Removed the @pytest.mark.flaky decorator from a chatbot streaming test, indicating increased confidence in its stability and reliability.

Removed

🗑️ Deprecated API Keys: Cleaned up the .env.dev file by removing AZURE_OPENAI_KEY_IMAGE, AZURE_OPENAI_KEY_AUDIO, and HUGGINGFACE_API_KEY, as these are now managed centrally or dynamically within the new configuration system.
🧹 Legacy Docker Compose Files: Eliminated redundant and outdated Docker Compose files, which are now replaced by the new dynamically generated configurations for improved maintainability.
📜 Unused License Entries: Updated licenses.config.json by removing entries for services no longer in use (e.g., rabbitmq, kafka, airflow, grafana, prometheus, and various base image references), ensuring an accurate and up-to-date record of third-party licenses.

[v0.252.2] - 2025-11-14 - Streamlined Docker Builds with Enhanced Security

Security

🔑 Enhanced Runtime Security: Significantly improved container security across all services by removing sudo installation and privileges for the non-root user, minimizing potential attack vectors.
🔒 Refined File Ownership: Configured Docker builds to ensure all application files copied to the runtime image are explicitly owned by the non-root user, preventing permission issues and strengthening security best practices.

Refactor

⚡️ Optimized Docker Image Layers: Consolidated multiple RUN commands into single instructions across all Dockerfiles, reducing the number of image layers for smaller image sizes and faster build times.
🧹 Trimmed Runtime Dependencies: Removed unnecessary system packages (tree, vim, iputils-ping, libasound2) from service runtime images, creating leaner and more efficient deployments.
⚙️ Standardized Entrypoint Configuration: Updated ENTRYPOINT definitions to the more robust "exec" form (["sh", "-c", "..."]) in Dockerfiles, enhancing command execution reliability and variable handling.
🎯 Improved Variable Quoting: Applied proper quoting for the POETRY_VERSION environment variable during Poetry installation commands, ensuring robustness and preventing potential parsing issues.
📁 Centralized Dagster Home Management: Standardized the DAGSTER_HOME directory for pipeline services to a dedicated, securely-owned location (/dagster_home), improving consistency and non-root user compatibility.
🛠️ Adjusted Build Tool Availability: Strategically included the make utility in the runtime images for aihub_api, aihub_bot, and aihub_pipeline services, ensuring necessary build tools are available only where required.
📦 Precise Environment PATH: Adjusted the PATH environment variable for the default_rag_pipeline to more accurately reference its virtual environment, enhancing environment isolation and reliability.

[v0.252.1] - 2025-11-14 - Strengthened Data Integrity and Pipeline Organization

Fixed

🐛 Corrected the description for the AIHubUser role, clarifying that it grants global user access instead of administrative access within AI-Hub.

Changed

🛡️ Implemented stricter validation and automatic sanitization for Namespace and Folder names in the RAG datalake. Names will now be restricted to alphanumeric characters, hyphens, and underscores, enhancing data integrity and system compatibility.
🔄 Restructured data pipeline asset keys by introducing clearer intermediate path segments (e.g., datalake_to_vectorstore, sharepoint_to_datalake), improving the organization and clarity of data ingestion and processing workflows.

Refactor

🧹 Streamlined internal dependencies by updating the import path for OpenAILike to directly reference llama_index.llms.openai_like.

[v0.252.0] - 2025-11-12 - Broadening Horizons: Local Filesystem Integration and Generalized Data Sources

Added

🚀 New Local Filesystem Ingestion Pipeline: Introduced a complete pipeline, resource, and I/O manager for seamlessly integrating and syncing files from local or network file systems to the data lake, supporting flexible regex-based filtering.
⚙️ Generic Source File Abstraction: Implemented a new SourceFile interface and generic factories/operations, enabling the pipeline to ingest data from any source system that adheres to this standard, making it highly extensible.
🏷️ Pytest flaky Marker: Added official flaky marker definition to pyproject.toml and applied it to unreliable tests, improving test suite clarity and stability by allowing flaky tests to be excluded from CI.
➕ Regex Pattern Utilities: Introduced helper functions for generating powerful regex patterns, used by the new local filesystem resource for flexible file and folder filtering.

Changed

🗓️ Standardized Source File Timestamps: Converted all file creation and modification timestamps in source file models (e.g., SharePoint) to Unix timestamps (int) for consistency and easier processing.
🔓 Relaxed Namespace Naming Constraints: Removed restrictive regex validation for NamespaceEntity names, allowing more flexibility in naming conventions.
📈 Enhanced Metadata Display: Updated metadata tables to reflect new data structures and added detailed metadata display for local filesystem files in the Dagster UI.
🔧 Pipeline Definition Parameters: Adjusted default_definitions to remove the explicit figures_directory_name parameter and introduced vector_store_dimensions and auto_sync for more flexible configuration.

Refactor

🗄️ Abstracted Data Lake Client Operations: Replaced specific data lake client implementations (Azure, S3) with an AbstractDataLakeClient interface, standardizing URI construction, file deletion, and directory operations across different storage backends.
♻️ Consolidated Source-to-Data Lake Logic: Replaced SharePoint-specific data lake transformation and cleanup factories/operations with new generic source_to_data_lake counterparts, reducing code duplication and improving maintainability.
🖼️ Simplified Document Loader Configuration: Streamlined DoclingLoader and DocumentIntelligenceLoader by removing the explicit figures_directory_name parameter, making figure directory handling internal and consistent.
🔗 Decoupled Azure Search Metadata Definitions: Removed Azure Search specific metadata field definitions, decoupling the core metadata handling from particular vector store implementations.

Removed

🗑️ Deprecated SharePoint-Specific Factories: Eliminated older factories dedicated solely to SharePoint-to-Data Lake transformations, superseded by the new generic source ingestion framework.
❌ Legacy SharePoint-Specific Operations: Removed individual operations for extracting content, metadata, and URIs specific to SharePoint files, which are now replaced by their generic SourceFile counterparts.

[v0.251.4] - 2025-11-11 - Azure Integration Overhaul and Infrastructure Simplification

Removed

🗑️ Infrastructure-as-Code (IaC) Module: The aihub_iac module, which provided Pulumi-based Azure infrastructure definitions, has been entirely removed from the project, streamlining deployment and management.
🗑️ Azure AI Search Vector Store Integration: Direct support for Azure AI Search as a vector store option for the RAG agent has been removed, simplifying the vector store ecosystem and reducing dependencies.
🗑️ Azure Blob Storage for Anonymous File Access: The Azure Blob Storage implementation for anonymous file access has been removed, standardizing anonymous file access across the platform to S3-compatible services.
🗑️ Azure Cosmos DB for Bot Credentials Setup: The setup process for Azure Bots no longer supports saving credentials to Azure Cosmos DB, exclusively using MongoDB-compatible storage (e.g., FerretDB).
🗑️ Legacy Azure Data Lake (ADLS2) Integration: Older integration components for Azure Data Lake (ADLS2) and associated Dagster resources have been removed, transitioning to a modern, connection string-based approach.
🗑️ Direct Azure Service Integration Clients: Several direct integration clients and settings for Azure services (e.g., Azure Cognitive Search, Blob Storage, Speech Service) have been removed from aihub_lib/infrastructure/azure, simplifying the core library.
🗑️ Azure-Specific Test Tags and Scenarios: @azure pytest markers and associated test scenarios have been removed from agent tests, decoupling the test suite from specific Azure infrastructure components.

Changed

🔄 Azure Document Intelligence Configuration: The configuration for Azure Document Intelligence now explicitly requires a service endpoint URL and an API key, providing more direct and transparent authentication.
🔄 Azure Data Lake Storage Authentication: Azure Data Lake Storage integration within pipelines now exclusively uses an explicit connection string for authentication, enhancing configuration clarity and security.
🔄 Default Vector Store in Pipeline Examples: The default and recommended vector store resource used in pipeline examples has shifted from Azure AI Search to Milvus, reflecting updated recommendations.
🔄 S3 Anonymous File Access Service Clarification: The S3 anonymous file access service now explicitly mentions SeaweedFS as a compatible backend, clarifying its S3-compatible nature.

Refactor

🧹 Python Generics Syntax Modernization: Updated generic type hint syntax across several NATS event and publisher/subscriber classes to align with Python 3.12+ features, improving type safety and code readability.
🧹 Centralized Azure Document Intelligence Client Access: Simplified the method for accessing the Document Intelligence client, replacing a custom singleton with direct instantiation using new explicit settings.
🧹 Unified Anonymous File Access Implementation: The internal implementation of the anonymous file access service has been unified to solely rely on S3-compatible storage, removing conditional logic for multiple cloud providers.
🧹 Agent Testing Infrastructure: The testing setup for agents has been refactored to remove Azure-specific configurations and dependencies, making tests more portable and less coupled to particular cloud providers.
🧹 Azure Configuration Reorganization: Azure-related settings (e.g., Document Intelligence, Data Lake) have been reorganized into more specific top-level modules within aihub_lib/infrastructure/, improving module clarity.
🧹 Workflow Visualizer Event Naming: Minor adjustment to the string slicing logic in the workflow visualizer for more accurate extraction of event names.

[v0.251.3] - 2025-11-11 - Enhanced Multilingual Documentation and Content Synchronization

Added

📄 Introduced Multilingual Code Deep Dive: Added dedicated English and German index pages for the new "Code Deep Dive" documentation section, which dynamically pulls READMEs, changelogs, and licenses directly from the codebase.
🌍 Expanded Multilingual Support for Core Documentation: Implemented automatic generation of both English and German versions for the changelog and license documentation, improving accessibility for diverse audiences.

Changed

⚡️ Optimized LLM Documentation Generation: Adjusted the process for generating documentation used by Large Language Models to exclude the "Code Deep Dive" section from the full content output (llms-full.txt), while retaining it within the table of contents.
🖼️ Improved Documentation Readability: Applied minor formatting and line break adjustments across several quick-start and architectural decision documents for enhanced clarity.

Refactor

🔄 Refined Documentation Synchronization Script: Updated the sync-docs.sh script to intelligently manage multilingual documentation, ensuring correct cleanup of synced files and automatic duplication of English content to German for the "Code Deep Dive" section.

[v0.251.2] - 2025-11-11 - Quick Start Guide Enhancements: Clearer Conversations and Deeper Insights

Added

🎬 New Video Tutorials: Introduced two new video tutorials in the "Your First Conversation" guide, visually demonstrating how to initiate conversations via the main chat interface and directly through an agent.

Changed

📄 Revamped "Your First Conversation" Guide: Significantly expanded and clarified the guide on starting conversations with agents, detailing two primary methods and highlighting key features of the chat interface (Open WebUI), along with options for model and agent selection.
📖 Enhanced "Your First Insights" Documentation: Updated the guide to provide comprehensive explanations of "Traceability" and "Document Retrieval (RAG)" features, illustrating how users can gain deeper insights into agent interactions directly from the chat interface.
📝 Refined "What Just Happened?" Section: Improved the introductory text and streamlined descriptions for navigation options in the quick start summary, enhancing overall readability and guidance.

[v0.251.1] - 2025-11-10 - Streamlined AI Assistant Integration via AGENTS.md Documentation

Added

🦾 Introduced Comprehensive AI Agent Developer Guides (AGENTS.md): New, in-depth documentation files have been added at the project root and within each core Python package (e.g., aihub_action, aihub_agent, aihub_api). These guides provide detailed context for AI assistants and developers on package purpose, tech stack, architecture, and development workflows.
⚙️ Configured Gemini AI Assistant Context: A new .gemini/settings.json file has been added to explicitly instruct the Gemini AI assistant to use the new AGENTS.md files for codebase context, ensuring more relevant assistance.

Changed

🔄 Updated Claude AI Assistant Context References: Existing CLAUDE.md configuration files across the repository have been updated to reference the new, more specific AGENTS.md documentation, enhancing Claude's understanding of each package's role and structure.

Removed

🗑️ Deprecated Generic Gemini Context Files: The GEMINI.md files, which previously pointed to generic README.md documentation, have been removed from the root and all packages. Their functionality is now superseded by the new .gemini/settings.json and dedicated AGENTS.md files.

[v0.251.0] - 2025-11-10 - NATS-Powered Eventing for Smarter Data Pipelines

Added

✨ Introduced Event-Driven Document Pipelines: Implemented new capabilities to automatically trigger data ingestion pipelines via NATS events when documents are uploaded and validated.
🚀 New NATS Poller Utility (JSPoller): Added a robust and generic JetStream poller, simplifying asynchronous event processing and message acknowledgment from NATS.
📄 SourceUpdatedEvent for Data Lake Changes: Introduced a dedicated NATS event type to signal when a file has been updated or uploaded in the data lake, enabling reactive workflows.
📦 Structured Pipeline Topic Managers: Added PipelineTopicManager and PipelineInstanceTopicManager to standardize NATS subject naming conventions for pipeline-related events, enhancing routing and observability.
🚦 NATS Document Upload Sensor for Dagster: Integrated a new Dagster sensor that listens for SourceUpdatedEvent messages on NATS, automatically initiating pipeline runs for newly uploaded documents.
🏷️ Customizable Dagster Schedule Names: Enabled the ability to assign explicit names to Dagster schedules, improving identification and management of periodic tasks.
🔗 NATS Endpoint Configuration for Dagster: Configured Dagster services to connect to NATS, enabling seamless event communication within the pipeline ecosystem.

Changed

⚡️ Knowledge Document Upload Validation: The API endpoint for validating document uploads now publishes a SourceUpdatedEvent to NATS upon successful verification, integrating with event-driven pipelines.

Refactor

🧹 Unified NATS Topic Manager Design: Refactored all TopicManager classes to inherit from Pydantic's BaseModel, enhancing type safety, consistency, and data validation across NATS subject generation.
🔄 Standardized Topic Manager Instantiation: Updated constructors for AgentClassTopicManager, AgentInstanceTopicManager, ProcessClassTopicManager, and related classes to use keyword arguments, improving code clarity.
🦾 Simplified JetStream Event Replay: Replaced manual JetStream consumer creation and pull subscription logic in JetStreamEventStore with the new JSPoller utility, streamlining historical event replay.
🛡️ Improved NATS Stream Management for Publishers: Enhanced JSPublisher with a new ensure_stream_exists method, allowing publishers to proactively verify and create necessary JetStream streams.
⚙️ Cleanup of observable_data_lake_factory: Removed an unused DataLakeResource parameter from the observable_data_lake_factory asset, simplifying its signature.
📚 Centralized Pipeline Constants: Moved internal pipeline source and target names (e.g., 'datalake', 'knowledge') into dedicated constant files for better maintainability and consistency.
💬 Clarified Process Topic Descriptions: Updated comments for process-related NATS topics to accurately reflect their purpose, improving documentation.

[v0.250.2] - 2025-11-07 - Streamlined Local Development and Documentation Clarity

Added

✨ New Local Development Option (Build from Source): Introduced docker-compose.build.yml to allow developers to build and run the entire platform locally directly from source code, facilitating active development and debugging.
🎤 Microphone Access for Open Web UI: Enabled microphone permissions within the Open Web UI iframe, supporting voice-based interactions and features.
📄 macOS mkcert Instructions: Added specific installation instructions for mkcert on macOS to the local development prerequisites, improving setup guidance.
📊 Deployment Comparison Summary: Included a new summary table highlighting the key differences between production and local deployment configurations and purposes, enhancing clarity for users.

Changed

🔄 Refined Local Deployment Strategy: Reworked local development guidance, clearly separating the use of docker-compose.local.yml for running with pre-built images (recommended for testing) from the new docker-compose.build.yml for building from source (for active development).
📄 Enhanced Documentation Structure: Significantly reorganized and updated the Prerequisites and One-Command Deployment documentation, providing clearer, distinct instructions and sections for production versus local deployments.
🔑 Local Traefik SSL Configuration: Modified the local Traefik setup (docker-compose.local.yml) to exclusively use self-signed SSL certificates via tls=true and dynamic file providers, removing the dependency on Let's Encrypt for local environments.
📋 Version-Specific Traefik Middlewares: Updated Traefik configurations for production and nightly builds to load middlewares from version-specific files, allowing for more granular control.
🛠️ Improved Audio Transcription Filenaming: Enhanced the internal logic for naming audio chunks during transcription to ensure more unique and descriptive filenames.
📝 Updated Architecture Decisions: The containerized deployment architecture documentation has been updated to reflect the new local development options and their corresponding Docker Compose files.

Removed

🗑️ Deprecated Central Traefik Middleware File: The standalone traefik/middlewares.yml file has been removed, as middleware configurations are now dynamically loaded from specific configs directories for better modularity.

[v0.250.1] - 2025-11-07 - Streamlined Deployment and Core Service Upgrades

Added

✨ Milvus UI Integration: Introduced Attu (Milvus UI), making vector database management more accessible via a dedicated subdomain (attu.your-domain.com).
🦾 New RAG Agent: Added a dedicated rag_agent service for enhanced Retrieval Augmented Generation capabilities, integrated with core platform services.
📄 Dagster Workspace Configuration: Introduced workspace.yaml files for Dagster, enabling dynamic pipeline loading from the default_rag_pipeline service.
🛡️ Default Security Headers: Implemented a new security-headers middleware in Traefik to enhance platform security by default.

Changed

⚙️ Refined Deployment Documentation: Significantly updated and expanded the English prerequisites and one-command deployment guides, including detailed DNS, Azure Entra ID token version 2 requirement, and comprehensive app role setup instructions.
🚀 Standardized Azure OpenAI Configuration: Consolidated Azure OpenAI API key management and base URL configuration within LiteLLM, simplifying environment setup by using a single AZURE_OPENAI_BASE_URL and AZURE_OPENAI_KEY.
📊 Updated OpenWebUI: Upgraded OpenWebUI to v0.6.28, bringing various improvements and bug fixes, and updated its internal API endpoints to align with the platform's authentication and data access patterns.
🧹 Improved Core Service Health Checks: Adjusted Docker Compose healthcheck configurations for Phoenix, NATS, Valkey, and Milvus UI (Attu) to prevent false-negative restarts and improve startup reliability.
🔑 Updated Authentication Configuration: Enhanced authentication parameters within the API and Bot services, ensuring consistent propagation of OAuth and superuser credentials.
⚙️ Enhanced Pipeline Configuration: Explicitly passed numerous environment variables to the default_rag_pipeline service, improving its configurability and robustness.
🖼️ Jupyter Lab Data Volume: Changed the default data volume for Jupyter Lab from llama-data to jupyter-data for clearer separation of concerns.
🔐 Traefik Dashboard Domain: Switched the Traefik dashboard host rule from traefik.localhost to traefik.${DOMAIN} for consistent domain usage across the platform.
🛠️ LiteLLM Guardrail Defaults: Changed default presidio-mask-guard and presidio-block-guard settings to default_on: false in LiteLLM configurations, allowing for explicit enablement.
📦 Docling Service Configuration: Updated the Docling image to v1.3.1 and refined its configuration with explicit artifact paths and model settings.

Fixed

🐛 Robust Makefile Sourcing: Modified aihub_bot/Makefile to optionally include .env files, preventing build failures when the file is absent.
🩹 Websocket Endpoint Correction: Corrected the WebSocket endpoint URL for frontend services from /api/v1/event/ws to /api/v1/events/ws (plural).
🔒 Docling Model Permissions: Introduced a new docling-model-permission service to correctly set ownership for model volumes, resolving potential permission issues.
🛠️ PostgreSQL Healthcheck User: Updated the PostgreSQL healthcheck to use the configured POSTGRES_USER variable for improved flexibility and consistency.

Removed

🗑️ German Documentation: Removed the German versions of the "Prerequisites" and "One-Command Deployment" documentation pages to streamline content.

Refactor

🔄 Centralized Traefik Configuration: Streamlined Traefik configuration by moving some parameters directly into the command section and standardizing SSL certificate resolution with Let's Encrypt.

[v0.250.0] - 2025-11-06 - Next-Gen Bot Framework: Microsoft Agents SDK Integration and Multi-Bot Capabilities

Added

✨ New Bot-in-the-Loop Start Event: Introduced BotInTheLoopAgentStartEvent to provide a more flexible and validated way to initiate Bot-in-the-Loop workflows, supporting explicit Teams or Slack configurations from the agent.
⚙️ Structured Slack Configuration: Added SlackConfig within BotInTheLoopRequestEvent to enable robust and explicit configuration details when initiating Slack interactions.
🧪 Comprehensive Testing Utilities: Introduced new testing fixtures for mocking MSAL authentication and aiohttp requests, significantly enhancing test isolation and reliability for aihub_bot components.

Changed

🚀 Upgraded Bot Framework SDK: Migrated aihub_bot components from the legacy botbuilder library to the new microsoft-agents SDK. This fundamental upgrade impacts bot hosting, authentication, activity handling, and overall integration with Microsoft Bot Framework channels.
🤝 Enhanced Multi-Bot Conversation Management: Conversation tracking and persistence (in ConversationEntity and ConversationTracker) now explicitly incorporate bot_id. This ensures robust data isolation and prevents collisions when multiple bot instances operate across various channels or environments.
👤 Improved Teams User Identity Handling: Enhanced the process of extracting user identity for Teams interactions, leading to more accurate retrieval of aad_object_id, email, and roles.
🤖 Refined Bot-in-the-Loop Logic: Updated the core logic for the BotInTheLoopAgent and its handler to seamlessly support multi-channel configurations (Teams and Slack) and streamline the sending of proactive messages using the new SDK.
📄 Documentation Updates: Minor updates to documentation source_sha values.

Fixed

🐛 NATS BaseEvent Serialization: Corrected an issue in BaseEvent serialization to ensure all model fields are properly dumped during the process.

Refactor

🧹 Streamlined Bot Activity Models: Removed the custom ActivityModel as the new microsoft-agents SDK provides direct, improved activity handling, simplifying controller definitions.
🔄 Centralized Bot Message Handling: Refactored common bot message processing logic into a shared _handle_message method, simplifying the distinction between direct and channel interactions for Teams and Slack.

[v0.249.4] - 2025-11-06 - Refined Document Metadata for Clearer Source Tracking

Added

✨ Introduced source_origin field: Documents and nodes now include a new source_origin field to explicitly store the original external URI (e.g., SharePoint URL) distinct from the data lake URI.

Changed

🔄 Clarified source field definition: The source metadata field for ingested documents and nodes is now explicitly defined as the data lake URI.
🚀 Improved document ingestion: Pipelines for ingesting documents (e.g., from SharePoint, data lake) have been updated to correctly capture and differentiate between the data lake URI (now source) and the new source_origin.

Removed

🗑️ Deprecated DATA_LAKE_URI metadata field: The DATA_LAKE_URI field has been removed to streamline and simplify document metadata, with its functionality now handled by the source field.

[v0.249.3] - 2025-11-05 - Streamlined LiteLLM API Parameter Handling

Changed

🔄 Improved LiteLLM Model Parameter Handling: Configured various LiteLLM models (e.g., text-generation/mini, text-generation/small, text-generation/large) across all environments (dev, latest, local, nightly) to automatically drop extraneous parameters from API requests, enhancing compatibility and robustness with underlying services.

[v0.249.2] - 2025-11-05 - Enhanced Model Agnosticism and New Azure GPT-5 Series Integration

Added

✨ New Azure OpenAI GPT-5 Nano Integration: Introduced direct support for Azure OpenAI GPT-5 Nano via the text-generation/nano alias, including a dedicated environment variable (AZURE_OPENAI_KEY_NANO) and specific default temperature settings for optimal performance.
🚀 New Azure OpenAI GPT-5 Series Integration: Expanded LLM capabilities with support for Azure OpenAI GPT-5 via the text-generation/large alias, providing access to more powerful generative models.

Changed

⚙️ Updated LiteLLM Proxy: The internal LiteLLM dependency has been updated to v1.77.7-stable, bringing general improvements and stability fixes to the model proxy layer.
🤖 Adjusted Default Fallback and Safety LLM Configuration: The system's default fallback LLM and the prompt injection detection LLM now leverage the newly introduced text-generation/nano model alias.

Refactor

🧹 Standardized Model Aliases Across the Platform: Implemented a consistent, provider-agnostic model naming scheme, replacing previous explicit model names (e.g., azure/gpt-4o-mini, local/qwen-embedding, local/reranker) with simplified aliases like text-generation/mini, embedding/small, reranker, transcription, speech, and image-generation. This refactoring significantly simplifies model configuration and enhances platform flexibility.
🔄 Unified Model Configuration: All agent definitions, API endpoints, infrastructure-as-code (IaC), and data pipeline configurations have been updated to utilize the new standardized model aliases, ensuring consistency and ease of maintenance.
📈 Enhanced LLM Parameter Configuration: Introduced capabilities for defining default parameters within LLMConfig, allowing for fine-grained control over model behavior for specific aliases, such as setting a default temperature for text-generation/nano.

[v0.249.1] - 2025-11-04 - Enhanced Authentication Service Robustness

Fixed

🐛 Resolved intermittent connectivity issues with external authentication and identity services (JWKS and Azure Graph API) by extending connection timeouts to better accommodate IPv6 fallback mechanisms.

[v0.249.0] - 2025-11-04 - Streamlined Knowledge Management and Enhanced Security

Added

✨ Introduced Document Upload Endpoints: Dedicated API endpoints are now available for initiating and validating document uploads directly within specific knowledge base databases and namespaces.
📄 New Endpoint for Supported File Types: A new API endpoint has been added to retrieve a list of supported file types for client-side validation of knowledge base document uploads.
🔑 Granular Permissions for Knowledge Base: Implemented fine-grained access control for knowledge base resources, allowing permissions to be defined per database and namespace for enhanced security.
🆔 Added database_id to Namespace DTO: The Namespace Data Transfer Object now includes a database_id to provide clearer data relationships and context.

Changed

🔄 Updated Knowledge Base API Routes: Namespace creation and update endpoints, along with document retrieval, now utilize database and namespace names directly as path parameters for improved clarity and consistency.
🛠️ Refined Database Listing for Users: The /databases endpoint now dynamically filters displayed databases and namespaces based on the requesting user's granular access permissions, ensuring only accessible resources are shown.
⚙️ Configured CORS for SeaweedFS: The SeaweedFS S3 gateway now automatically configures Cross-Origin Resource Sharing (CORS) rules during initialization, improving compatibility for direct client-side file uploads.
🌐 Web Client Adaptation: The web application's document management components have been updated to seamlessly integrate with the new knowledge base API structure, including the revised document upload and namespace management flows.

Removed

🗑️ Deprecated Generic File Upload Endpoints: Removed the standalone, generic file upload and validation endpoints from the /files API, centralizing all document-related uploads under the Knowledge Base domain for better organization.

Refactor

🧹 Centralized Document Upload Logic: Consolidated all document upload and validation business logic and associated Data Transfer Objects (DTOs) from the generic FileService to the specialized KnowledgeService, enhancing modularity and maintainability.
♻️ Renamed File Upload DTOs: Data Transfer Objects related to file uploads (e.g., FileUploadRequest to DocumentUploadRequest) were renamed to better reflect their new purpose and scope within the Knowledge Base context.

[v0.248.1] - 2025-11-03 - Enhanced Document Processing and Data Lake Organization

Changed

🖼️ Knowledge Interface Clarity: The knowledge interface now automatically excludes generated figure files (artifacts from document processing) from display, providing a cleaner and more focused view of core documents.

Refactor

🧹 Centralized Figure Directory Management: The __figures__ directory name used for storing document figures has been centralized as a global constant, significantly improving consistency and maintainability across the system.
🔄 Simplified Data Lake Operations: Streamlined data lake client methods and operations by removing the explicit figures_directory_name parameter, as the directory name is now consistently managed internally.
⚙️ Optimized Document Loaders: Updated DoclingLoader and DocumentIntelligenceLoader to utilize the centralized figure directory constant, simplifying their API and ensuring consistent figure path generation.
🛠️ Refined Resource and Pipeline Definitions: Cleaned up various resource and pipeline definitions, including DataLakeResource and associated factories, by removing redundant figures_directory_name parameters.

Removed

🗑️ SonarLint Configuration: Deleted an obsolete SonarLint project configuration file from the codebase.

[v0.248.0] - 2025-11-03 - Smarter Document Understanding: Advanced Table Processing

Added

✨ Intelligent Table Splitting: Introduced the capability to automatically split large tables into smaller, manageable chunks, ensuring better retrieval and processing of tabular data while respecting token limits.
🦾 LLM-Powered Table Header Detection: Implemented LLM-driven analysis to accurately identify multi-row table headers, improving the precision of table parsing and preserving hierarchical structures during splitting.
🧱 New Dependency lxml: Added lxml to enhance HTML parsing and manipulation capabilities, foundational for advanced table processing.

Changed

🔄 Advanced Table Conversion in DoclingLoader: Modified DoclingLoader to now convert Markdown tables into structured HTML tables, improving their representation and downstream processing.
⚡️ Optimized Document Intelligence Table Handling: Streamlined table processing in DocumentIntelligenceLoader by removing redundant table reformatting, leveraging the new unified table parsing logic.
📄 Improved Markdown Table Parsing: Enhanced MarkdownStructuralNodeParser to intelligently process and chunk tables, ensuring headers are preserved and content fits within token limits for improved Retrieval-Augmented Generation (RAG) performance.
🚀 Increased Agent Test Timeout: Extended the delay_before_stop parameter in Retrieval Agent tests to provide more time for operations to complete, improving test stability.

Removed

🗑️ Deprecated NODE_CONTENT_TYPE_TABLE: Removed the NODE_CONTENT_TYPE_TABLE metadata tag, as table content is now directly represented within HTML structures, simplifying content type management.
🧹 Eliminated Redundant Table Reformatting: The dedicated reformat_tables function in DocumentIntelligenceLoader has been removed, as its functionality is now integrated and optimized within the core parsing mechanisms.

[v0.247.7] - 2025-11-03 - Enhanced RAG with Hybrid Search and Model Updates

Added

🚀 Introduced Hybrid Search Capability: Enabled Milvus vector stores to perform hybrid search queries by integrating sparse embeddings and BM25 functionality, improving retrieval relevance.

Changed

🔄 Default Query Mode to Hybrid: Updated the default query mode for RAG and Retrieval agents to HYBRID, leveraging the newly added hybrid search capabilities in Milvus for more comprehensive retrieval.
⬆️ Updated Default LLM Model: Switched the default Large Language Model in RAG pipelines from gemma-3-multimodal-small to qwen-2.5-multimodal-small, enhancing model performance and capabilities.

[v0.247.6] - 2025-10-31 - Empowering Contributors with New Guidelines

Added

✨ Added a comprehensive CONTRIBUTING.md guide: This new document provides clear instructions and best practices for anyone looking to contribute to the Swiss AI-Hub project, covering everything from code contributions to documentation, community engagement, and advocacy, fostering a more collaborative environment.

[v0.247.5] - 2025-10-31 - Enhanced Security Measures

Security

🔑 Established comprehensive security policy: Introduced a new SECURITY.md file, providing clear guidelines on how to report security vulnerabilities, outlining supported versions, and detailing the coordinated disclosure process.

[v0.247.4] - 2025-10-31 - Community Guidelines Established

Added

📄 Introduced Code of Conduct: Adopted the Contributor Covenant 3.0 to foster a welcoming, safe, and equitable community by outlining expected behaviors, restricted actions, and a clear process for reporting and addressing issues.

[v0.247.3] - 2025-10-31 - Empowering Bots with Teams Integration and Enhanced Core Functionality

Added

✨ Microsoft Teams Support for Bot-in-the-Loop: Expanded the Bot-in-the-Loop pattern to support Microsoft Teams, enabling agents to request human input directly within Teams channels.
🚀 New Makefile Targets: Introduced up-dev for easy development environment startup and generate-api-token for simplified API token generation. Also added an agents-playground target for starting Agents Playground with Azure OpenAI configuration.
⚙️ CloudAdapter Caching: Implemented caching for CloudAdapter instances to improve performance and reduce repeated authentication overhead in bot routes.
📄 Bot-in-the-Loop Architectural Documentation: Added a comprehensive explanation to the README.md detailing the "Handler vs Bot" separation of concerns for the Bot-in-the-Loop pattern.
👤 TeamsConfig Model: Introduced a TeamsConfig Pydantic model in aihub_lib to standardize configuration for Microsoft Teams interactions within Bot-in-the-Loop requests.
🔑 Dynamic OpenAI Client Acquisition: Integrated LiteLLMService to dynamically acquire the OpenAI client based on user identity, enhancing security and flexibility for chat completions.

Changed

🔄 Bot-in-the-Loop Agent Example: Updated the BotInTheLoopAgent playground example to demonstrate the new TeamsConfig for multi-channel communication.
💬 Enhanced Content Extraction for Teams: Improved the ContentExtractor to correctly handle HTML content and emojis from Microsoft Teams messages, providing richer context to AI models.
⚙️ Centralized Chat Completion Logic: Refactored the AgentChatController and OpenaiChatController to centralize common request processing logic, improving code maintainability and consistency.
📄 Generalization of Responder Information: Renamed SlackResponderInfo to BotInTheLoopResponderInfo and added aad_object_id to support generic user information across multiple channels like Slack and Teams.
🚀 Bot-in-the-Loop Thread Identification: Generalized thread identification in BotInTheLoopHandler to use thread_identifier instead of slack_thread_ts, accommodating multi-channel requirements.

Fixed

🐛 Robust Tracing Output: Ensured AgentRunTracer correctly serializes all output events by adding default=str to json.dumps, preventing serialization errors during tracing.
🐛 Teams Conversation Deletion Logic: Corrected the logic for deleting Teams conversations to prevent unintended deletions when a bot is added to a new team or channel.
🐛 Content Extractor Attachment Handling: Improved the ContentExtractor to intelligently skip non-file HTML content and emoji image attachments in Teams, preventing processing errors and focusing on relevant files.
🐛 Bot-in-the-Loop Handler Path: Corrected the base path for BotInTheLoopHandler to align with the /api/v1 endpoint structure.

Refactor

🧹 Conversation Entity Management: Renamed delete_conversation to delete_conversation_if_exists in ConversationEntity and added existence checks for more robust management.
🧹 Internal Method Naming Consistency: Standardized the naming convention of internal Slack-specific and conversation management methods to private (e.g., _is_slack_channel_message, _add_messages_to_conversation) for better encapsulation.

Removed

🗑️ Deprecated OpenAI Client Parameter: Eliminated the direct client parameter from OpenaiChatBot and its completion methods, as client acquisition is now handled internally by OpenaiCompletionHandler using the new LiteLLMService integration.

[v0.247.2] - 2025-10-31 - Optimized CI Documentation Builds

Changed

🚀 Streamlined CI documentation deployment: Updated the GitHub Actions workflow for documentation deployment to utilize a new, optimized build script, enhancing efficiency and reducing build times in continuous integration environments.
⚙️ Introduced dedicated docs:build:ci script: Added a specialized docs:build:ci command to package.json, which supports the faster documentation building process for CI by omitting the translation step.

[v0.247.1] - 2025-10-30 - Enhanced Documentation System and LLM Control

Added

⚙️ Automated Documentation Sync and Translation: Implemented a new GitHub Actions workflow (sync-docs) to automatically synchronize and translate documentation during the release process, ensuring documentation is always up-to-date and available in multiple languages.
🤖 LLM-Digestible Documentation Generation: Introduced a new build step that generates consolidated, LLM-optimized Markdown files (llms.txt, llms-full.txt) for each language, enhancing the ability of internal AI tools and coding assistants to leverage platform documentation.
🇩🇪 Expanded German Documentation Coverage: Significantly increased the amount of available documentation in German across various platform features, security, and compliance topics.
🧰 Markdown Copy/Download Buttons: Integrated vitepress-plugin-llms to add convenient copy and download buttons to Markdown code blocks within the documentation, improving usability for developers.

Changed

🔄 Documentation Build System Overhaul: Refactored the core documentation build process, externalizing sidebar generation logic to a dedicated module and integrating new plugins for improved maintainability and functionality.
🚀 Documentation Deployment Trigger: Modified the documentation deployment workflow to be triggered by a release-ready event from the main release pipeline, ensuring docs are deployed in sync with new software versions.
💰 LLM Proxy Cost Control Configuration: Activated and documented per-user budget, rate limiting (TPM, RPM), and parallel request limits for the LLM proxy, allowing administrators to configure and enforce spending controls via environment variables.
🌐 Improved German Translation Guidelines: Updated translation instructions for LLM models to better handle technical terms, avoid over-translation, and correctly manage absolute and relative links, leading to higher quality German documentation.
📝 Streamlined RAG Agent Documentation: Simplified and made more concise the explanation of the RAG Agent's principles and functionality in the English documentation.

Fixed

🔗 Documentation Internal Link Resolution: Corrected relative and absolute internal links within various compliance and platform documentation pages to ensure accurate navigation across the multi-language site.
🧹 Documentation Sync Script Exclusions: Enhanced the documentation synchronization script to exclude additional irrelevant files (e.g., __pycache__, .mypy_cache) and prevent the creation of orphaned directories from deeply nested README.md files.

Refactor

📦 Modularized Sidebar Logic: Moved the complex sidebar generation logic from config.mts to a new, dedicated ES module (sidebar-logic.mjs), improving modularity and maintainability of the documentation configuration.

Removed

🗑️ Outdated Deployment Documentation: Removed redundant and outdated documentation related to OpenStack deployment and general code deep dive sections that no longer reflected current practices.
🗑️ Deprecated RAG Pipeline Explanation: Eliminated a specific RAG pipeline explanation (docs/2_platform/6_pipelines/rag/index.en.md) as its content has been integrated or superseded by other documentation.

[v0.247.0] - 2025-10-29 - RAG Reranking and AI Model Updates

Added

✨ RAG Document Reranking: Introduced a new, configurable step in the RAG agent workflow to re-rank retrieved documents using a dedicated reranking model (local/reranker). This significantly enhances the relevance of information provided to the LLM, leading to more accurate answers.
🦾 Reranking Infrastructure: Added llama-cpp-reranker as a new Docker Compose service to support on-premise reranking model inference, enabling efficient and localized document re-scoring.
⚙️ Reranking Configuration: Implemented a new RerankingConfig to allow users to easily enable/disable reranking and specify the reranking model and top_n results.
📄 Reranking Documentation: Updated the RAG Agent documentation to explain the new "Re-ranking for Relevance" step in the RAG Agent's workflow.
🧪 Reranking Test Scenarios: Added new test scenarios and BDD steps to validate the reranking functionality, ensuring its correctness and reliability.
🌐 Reranking Localization: Included new localization keys (reranking_results) across all supported languages for enhanced user experience during the reranking step.
🔄 LlamaIndex Node Conversion: Added a utility to seamlessly convert internal IngestedNode objects to LlamaIndex NodeWithScore objects, facilitating integration with LlamaIndex-based reranking.

Changed

⚡️ Updated Default AI Models: Replaced local/qwen3-small and local/gemma-3-multimodal-small with local/qwen-2.5-multimodal-small as the default chat model in playground examples and infrastructure configurations, improving multimodal capabilities.
⚙️ RAG Agent Workflow: Modified the order_nodes_by_documents_step in the RAG Agent to process nodes after potential reranking, ensuring optimal context ordering.
⬆️ LiteLLM Version: Upgraded the LiteLLM proxy to version v1.77.7-stable, bringing the latest features and stability improvements.
📈 Increased Token Context Window: Adjusted the llama-cpp service configuration to support a larger context window (from 8196 to 16384 tokens) and increased max output tokens (from 4096 to 8196), enabling the processing of longer inputs and generation of more comprehensive responses.
⚙️ LiteLLM Guardrail Settings: Modified LiteLLM configurations to disable heuristic and similarity checks for prompt injection by default, while keeping LLM API-based checks enabled for robust security.
📊 RAG Agent Retrieval Settings: Updated RAG Agent playground configuration to use index_namespaces=["simple"] and increased retrieve_k from 5 to 20 to fetch more diverse initial results for reranking.
🩺 Healthcheck Commands: Switched wget to curl in several Docker Compose healthcheck commands for improved compatibility and robustness.
📄 Documentation Refinements: Applied minor formatting and content adjustments across various platform documentation pages to improve readability and consistency.

[v0.246.12] - 2025-10-28 - Major Documentation Overhaul and Core Infrastructure Alignment

Added

✨ Introduced Multi-Language Documentation Support: The documentation now fully supports English and German, with a new automated translation pipeline ensuring content consistency (.en.md and .de.md files, and translate-docs.sh script).
📄 Expanded Platform Documentation: Significant new sections have been added across the platform, covering:
- Chat Interface functionality, observability, and strategic rationale.
- Access Management details, including comprehensive RBAC implementation guidance.
- Auditing & Observability with deep dives into OpenTelemetry and tracing.
- Language Models proxy, anonymization, and input/output guards.
- Cost Control mechanisms and optimization strategies.
- Slack & Teams Integrations concepts and setup.
- API Interfaces: OpenAI-compatible REST, Agent Interaction REST, WebSocket, and Model Context Protocol (MCP) servers.
- Security: Detailed documentation on Authentication, Input Validation, Container Security, Network Security, and Data Encryption.
- Compliance: Extensive coverage of Data Retention, GDPR, Swiss DSG, EU AI Act, Internationalization, and Data Subject Access Requests (DSAR).
- User Interface overview with agent, thread, knowledge, process, and evaluation management.
- Agent Fundamentals, RAG Agent mechanics, and Expert Asking Agent collaboration patterns.
- Pipeline Fundamentals and the RAG Ingestion Pipeline.
- Agentic Processes concepts.
- Knowledge Management organization through namespaces.
- Agent Evaluations for testing and measuring AI agent quality.
- External System Integrations approaches.
📚 New Internal Command for Solution Documentation: Added a Claude command (document-solution.md) to streamline the creation and editing of high-level solution concepts.
⚙️ Added Documentation Merger Script: A new combine_docs.py script facilitates merging German markdown files from specific directories into a single DOCX document.
🔑 Ensured pipx Availability for LLM Tools: The deployment workflow now explicitly installs pipx to manage llm and llm-gemini for automated documentation translation.
🚀 Integrated Gemini for Automated Documentation Translation: The CI/CD process now leverages the llm command with the llm-gemini plugin for automated translation of documentation.
💡 Introduced encodings.xml for IntelliJ Project Settings: Added a new .idea/encodings.xml file to specify character encoding for certain project files.
☁️ New Documentation for Openstack Deployment: Added a new section for setting up and deploying to Openstack (6_code_deep_dive/deployment/index.en.md).

Changed

🔄 Updated Core Infrastructure Components: Object storage was migrated from MinIO to SeaweedFS, the NoSQL database from MongoDB to FerretDB, and the in-memory cache from Redis to Valkey across the platform and relevant documentation.
📖 Reworked Documentation Structure for Multi-Language Support: Implemented a new directory structure (renaming .md files to .en.md) and updated VitePress configuration (.vitepress/config.mts) to support robust multi-language documentation, including dynamic sidebar generation and content rewrites.
🗺️ Enhanced Sidebar Sorting Logic: Improved documentation sidebar generation to sort entries numerically by prefix and then alphabetically, ensuring a more intuitive navigation experience.
🔒 Updated openssl command for Key Generation: Changed the quick start guide to use openssl rand -hex 32 for generating secure random strings, enhancing clarity.
✅ Refined CI/CD Documentation Build Process: The deploy-docs.yml workflow now explicitly includes an automated translation step before the main documentation build.
📦 Updated packageManager Version: The package.json now specifies pnpm@10.19.0 for package management.
🤖 Adjusted Claude Command Paths: Updated the internal document-feature.md Claude command to reflect the new documentation structure for SDK features (e.g., 3_sdk/1_feature_overview/ and index.en.md).

Removed

🗑️ Deprecated Legacy Feature Overview Sections: Removed outdated or refactored feature overview documents and their respective files (e.g., 2_platform/5_feature_overview/* and 3_sdk/6_feature_overview/expert-agents/index.md), as content has been integrated into new, more granular sections.
🗑️ Removed Redundant Security Model Documentation: The dedicated 2_platform/2_architecture/4_security_model/index.md file was removed, as its content is now covered by the expanded security documentation.
🗑️ Retired Platform Customization Sections: Several older customization documents (2_platform/4_customization/*) were removed as part of the broader documentation overhaul and refactoring into more specific areas.

[v0.246.11] - 2025-10-27 - AI-Hub SDK Documentation: Clarity and Depth for Agent and Pipeline Development

Added

✨ New Pipeline Documentation Structure: Introduced dedicated sections for Pipeline Fundamentals, Core Pipeline Patterns, and Data Ingestion Pipeline, providing a more structured learning path for building robust data workflows. These new guides cover essential concepts like assets, I/O managers, resources, and the two-stage ingestion architecture.

Changed

📄 Expanded Agent SDK Guides: Significantly enhanced and reorganized documentation for Quick Start, Agent Fundamentals, Core Workflow Patterns, Human in the Loop, Multi-Agent Systems, and Testing and Debugging. These updates offer greater clarity, detailed explanations, new mermaid diagrams, and practical code examples to guide agent development.
🚀 Improved Pipeline Automation Guide: Rewrote the Job Scheduling documentation to clarify the hybrid automation strategy, emphasizing observable-driven and change-triggered processing for efficient pipeline automation.
🖼️ Refined SDK Overviews: Updated the main Building Agents and Building Pipelines overview pages to introduce key principles, simplified quick-start examples, and a more intuitive learning progression for new and experienced developers.
⚡️ Minor Documentation Enhancements: Updated the Production Deployment and Agent Observation documentation for title consistency and minor content additions.

Removed

🗑️ Deprecated Pipeline Documentation: Removed outdated documentation pages for Pipeline Patterns and Data Ingestion Pipeline, replacing them with the new, restructured guides for improved clarity and accuracy.

[v0.246.10] - 2025-10-24 - Enhanced Security and Access Control for Data Lake and Workflows

Added

🔑 New OAuth Configuration Options: Introduced OAUTH_TENANT_ID and OAUTH_COOKIE_SECRET environment variables to provide more comprehensive control over OAuth2 authentication setups.
🔐 Secure Access for SeaweedFS Filer: Implemented a new oauth2-proxy service to enforce secure, authenticated access to the SeaweedFS Filer UI, safeguarding data lake management.
👥 Group-Based Access for SeaweedFS: Added SEAWEEDFS_OAUTH_ALLOWED_GROUPS environment variable, enabling specific group-based access control for the SeaweedFS Filer UI.
📄 Expanded OAuth Scopes for Proxies: Configured openid, email, and profile OAuth scopes for both Dagster and SeaweedFS authentication proxies, enhancing user information retrieval during login.
➡️ Configured OAuth Redirect URLs: Added explicit redirect URLs for both Dagster and SeaweedFS OAuth proxies, ensuring correct post-authentication routing.
🎨 Customizable Sign-in Logo: Enabled support for a custom sign-in logo (OAUTH2_PROXY_CUSTOM_SIGN_IN_LOGO) for OAuth2 proxies in production and nightly environments, allowing for branded user experiences.

Changed

🔄 Centralized SeaweedFS Filer Access Routing: All direct access to the SeaweedFS Filer UI is now routed through the newly introduced oauth2-proxy service, ensuring a consistent and secure authentication layer.
📝 Renamed Dagster OAuth Proxy: The generic oauth2proxy service dedicated to Dagster has been renamed to oauth2proxy-dagster for clearer identification and separation of concerns.
🛡️ Standardized Default Access Groups: Updated the default allowed group for Dagster's OAuth proxy to AIHubDeveloper, standardizing access control with the SEAWEEDFS_OAUTH_ALLOWED_GROUPS variable and aligning with common roles.

Refactor

🧹 Refined OAuth Proxy Architecture: Streamlined the deployment architecture by separating OAuth2 proxy configurations into distinct services for Dagster and SeaweedFS, improving modularity and maintainability.

[v0.246.9] - 2025-10-22 - Streamlined DoclingLoader File Handling

Refactor

⚡️ Optimized DoclingLoader file content retrieval: Improved the internal efficiency of DoclingLoader by switching to fs.cat_file for more direct and potentially faster reading of file contents.

[v0.246.8] - 2025-10-21 - Core Bot Service Setup and Infrastructure Expansion

Added

🚀 Introduced a dedicated Bot service to the core infrastructure, laying the groundwork for new bot functionalities and integrations.
🌐 Configured Traefik routing for the new Bot service, making its API accessible via bot.<DOMAIN>/api/v1 endpoints across all deployment environments.
✨ Added essential environment variables for the Bot service, including development-only fake authentication settings, to streamline local and development deployments.
🦾 Integrated a development-only authentication handler into the Bot API's core controllers (HealthController and OpenaiChatController), simplifying local testing and development.
📦 Included Gunicorn as the production web server for the Bot service, optimizing its runtime performance and stability.

Refactor

🧹 Corrected module import paths within the bot service's Makefile for improved consistency and robustness in production builds.

[v0.246.7] - 2025-10-20 - Performance Optimizations and SeaweedFS Development Enhancements

Added

🚀 Implemented LiteLLM model info caching: Introduced an LRU cache for LiteLLM model information retrieval, which significantly reduces redundant API calls and improves the performance of model lookups.
⚙️ Enhanced SeaweedFS S3 gateway environment configuration: Added specific environment variables for S3_PORT, S3_FILER, and S3_BIND_IP, providing more granular control over the S3 gateway service in the development environment.
➕ Exposed SeaweedFS internal ports: Explicitly exposed SeaweedFS master, volume, and filer gRPC and HTTP ports in docker-compose-gpu.dev.yml for improved diagnostics and integration.

Changed

⚡️ Improved asynchronous document loading performance: Refactored the DoclingLoader to offload synchronous file reading and response processing to a dedicated thread, preventing event loop blocking and enhancing performance during large document ingestions.
🔄 Centralized S3-compatible storage to SeaweedFS in development: Updated milvus and attu services in docker-compose-gpu.dev.yml to depend on the seaweedfs-s3 gateway instead of a separate MinIO instance, streamlining the local object storage setup.
💾 Optimized SeaweedFS volume configuration: Adjusted seaweedfs-master settings to use a volumeSizeLimitMB of 512MB and disabled volumePreallocate for more efficient resource utilization in development.

[v0.246.6] - 2025-10-15 - Improved CI/CD Tagging and Release Process

Added

🚀 Nightly Release Tagging: Introduced automatic creation and pushing of a nightly git tag during the release process, which points to the same commit as the versioned release, streamlining access to the very latest builds.
⚡️ Dedicated Docker Tagging Workflow: Added a new retag-docker job to the set-latest workflow, allowing for more specialized and separated management of Docker image latest tags.
🛡️ Source Tag Validation: Implemented a robust check within the set-latest workflow to verify that the source tag exists before attempting to move the latest git tag, preventing potential errors during automated tagging.

Refactor

🔄 CI/CD Tagging Workflow Modularity: Refactored the set-latest workflow by renaming the primary job to retag-git and introducing a new retag-docker job, significantly enhancing modularity and clarity in automated tagging processes.
⚙️ Streamlined Git Tag Operations: Updated the retag-git workflow with necessary contents: write permissions, a dedicated actions/checkout step, and bot user configuration to ensure secure, reliable, and explicit git tag manipulation.

[v0.246.5] - 2025-10-14 - Refined S3 URI Handling in Data Lake Components

Refactor

🧹 Standardized S3 URI Parsing: Replaced manual string slicing with the more robust str.removeprefix() method for parsing S3 URIs across S3DataLakeIOManager and S3DataLakeClient, enhancing reliability and code clarity.
⚡️ Improved S3 Path Normalization: Added lstrip("/") during S3 URI processing to ensure more consistent and robust path handling within the S3DataLakeIOManager.
📄 Enhanced S3 Write Logging: Updated the logging message for successful S3 writes to display the fully reconstructed S3 URI, improving clarity and debuggability.

[v0.246.4] - 2025-10-13 - S3 Writing Efficiency and Document Parsing Simplification

Changed

🔄 Default Document Parser Loader: The default document parsing loader in DocumentParserResource has been updated from a 'both' approach (combining Docling and Document Intelligence) to exclusively use Docling, providing a more predictable out-of-the-box experience.

Removed

🗑️ 'Both' Loader Option for Document Parser: The BOTH loader type option has been removed from DocumentParserResource, streamlining the configuration for document parsing to either DOCLING or DOCUMENT_INTELLIGENCE.

Refactor

⚡️ Optimized S3 Object Writing: The S3DataLakeIOManager now writes S3 objects using a single put_object call, combining content, metadata, and content type settings for improved efficiency and atomic operations.

[v0.246.3] - 2025-10-13 - Enhanced S3 URI Robustness

Changed

🚀 Improved S3 URI Parsing: The S3DataLakeClient now intelligently handles S3 URIs, automatically prepending the "s3://" prefix when a URI starts directly with the configured container name, reducing errors and making input more flexible.

[v0.246.2] - 2025-10-10 - Simplified Pipelines, Smarter SharePoint Sync

Added

✨ SharePoint Integration: Introduced new SharePointSettings for externalizing SharePoint configuration and a new factory (default_sharepoint_to_datalake_definitions) for easily setting up SharePoint-to-DataLake pipelines.
🦾 Standardized Data Lake Pipelines: Added a default_definitions utility that provides a highly configurable and centralized way to define DataLake-to-VectorStore pipelines, promoting reusability and reducing boilerplate.
⚡️ Flexible Data Lake Directory Management: The DataLakeResource and related URI extraction logic now support an optional directory name, improving flexibility for data organization within containers.
⚙️ Configurable Bucket Auto-Sync: Enhanced bucket creation to allow explicit control over the auto_sync property for BucketEntity instances.

Changed

🔄 RAG Agent Defaults: Updated the default RAG agent configuration to use a default index namespace and defaultknowledge collection name, streamlining out-of-the-box usage.
🔑 Externalized SharePoint Settings: SharePoint authentication details are now securely loaded from environment variables via SharePointSettings rather than being configured directly on the SharePointResource.
📄 SharePoint File Type Filtering: Modified the default SharePointResource behavior to include all file types if no specific patterns are provided, offering more permissive syncing by default.
🧹 Simplified Data Lake IO Manager Configuration: Streamlined the prefix generation for Azure Data Lake IO managers, making configuration more consistent with container names.

Refactor

🏗️ Centralized Pipeline Definitions: Consolidated the definition logic for RAG pipelines into definitions_util.py, replacing scattered, repetitive configurations in default_rag_pipeline and playground applications.
⚡️ Streamlined Resource Factories: Refactored data lake and LLM resource factories to support more flexible configurations and direct instantiation, reducing complexity.

Removed

🗑️ Deprecated Data Lake Container Resource: Eliminated the DataLakeContainerResource in favor of a more flexible DataLakeResource and consolidated pipeline definitions.
🗑️ Redundant Pipeline Boilerplate: Removed verbose, repetitive pipeline definitions from app/default_rag_pipeline/__init__.py and aihub_pipeline/playground/__init__.py by leveraging the new centralized definition utility.

[v0.246.1] - 2025-10-09 - Retrieval Agent Context Customization

Added

✨ Configurable Context Prompt for Retrieval Agent: Introduced a new context_prompt configuration option for the RetrievalAgent, allowing users to define a custom prompt that dictates how combined and ordered retrieved nodes are formatted.

[v0.246.0] - 2025-10-07 - Modernizing Data Persistence: Introducing SeaweedFS, FerretDB, and Valkey

Added

✨ SeaweedFS for Scalable Object Storage: Integrated SeaweedFS as the new S3-compatible distributed file system, providing robust and scalable object storage across all deployment environments.
💾 FerretDB as MongoDB-compatible Database: Introduced FerretDB, a high-performance, MongoDB-compatible database that leverages PostgreSQL as its backend, enhancing data flexibility and open-source alignment.
⚡️ Valkey for High-Performance Caching: Replaced Redis with Valkey, a Redis-compatible fork, for efficient in-memory caching and faster data retrieval.
🛡️ Automatic S3 Bucket Configuration: Enhanced the S3 data lake client to automatically ensure the existence of necessary buckets and configure Cross-Origin Resource Sharing (CORS) for seamless web access.
🚀 Docker Compose Services for New Stack: Added comprehensive Docker Compose services and initialization scripts to manage the lifecycle of SeaweedFS (master, volume, filer, S3 gateway), FerretDB (and its PostgreSQL backend), and Valkey.

Changed

🔄 Updated Core Data & Storage Stack: Migrated the entire core infrastructure from MinIO, MongoDB, and Redis to SeaweedFS, FerretDB, and Valkey respectively.
📄 Documentation and README Updates: Revised README.md and quick start guides to reflect the new data persistence and caching technologies, including updated environment variables, service descriptions, and console links.
🔗 Service Integration Updates: Adjusted internal dependencies and connection strings for critical services like Milvus, OpenWebUI, and AI-Hub pipelines to seamlessly integrate with the new storage and database components.
✅ Improved Licensing for Core Services: Switched from AGPL/SSPL-licensed components (MinIO, MongoDB, Redis) to more permissively licensed alternatives (Apache-2.0 for SeaweedFS/FerretDB, BSD-3-Clause for Valkey), enhancing open-source compliance.

Removed

🗑️ MinIO S3 Storage Stack: The entire MinIO setup, including its Docker Compose services, entrypoint scripts, and related configurations, has been deprecated and removed.
❌ MongoDB Database: The MongoDB service and all associated configurations have been entirely replaced and removed from the core infrastructure.
🚫 Redis In-Memory Cache: The Redis service has been replaced by Valkey and removed from all deployment configurations.

[v0.245.9] - 2025-10-07 - Enhanced User Data Handling in Authentication

Changed

⚡️ Improved User Header Encoding: Switched from Base64 to URL encoding for usernames in authentication headers and signatures, enhancing compatibility and robustness when handling special characters.

[v0.245.8] - 2025-10-07 - Streamlined Data Lake Operations and Document Management

Added

✨ New RefDoc.get_documents method: Introduced a more flexible way to fetch reference documents without requiring a namespace filter, enhancing document retrieval capabilities.

Changed

🔄 Namespace-Agnostic Document Removal: The document removal process in data lake pipelines now uses the new get_documents method, making it more flexible and not bound to a specific namespace.
⚡️ Enhanced MongoDB Connection Management: Implemented explicit disconnect() calls in finally blocks across key utility functions and pipeline definitions, ensuring robust database connection cleanup after operations.
🧹 Refined Bucket and Namespace Utilities: The get_db_name_from_bucket_name and get_or_create_namespace_for_directory functions now include an auto_sync parameter, providing clearer differentiation for autoloading versus manual data ingestion pipelines.
🚀 Simplified Pipeline Configurations: Streamlined parameter passing for bucket and namespace information in default RAG and playground pipelines, allowing for direct use of container names and reducing redundant configuration.

Refactor

🛠️ Centralized Bucket and Namespace Logic: Extracted and consolidated core bucket and namespace creation/retrieval logic into new internal helper functions within bucket_utils, improving code organization and reusability.
💡 Optimized Store Name Retrieval: Removed an unnecessary wrapper function for retrieving the store name in the default RAG pipeline, leading to more direct and efficient utilization of the bucket_utils functionality.

[v0.245.7] - 2025-10-02 - Agent Guard Message Consistency

Fixed

🐛 Corrected RAGAgent guard rejection messaging: Ensured the RAGAgent accurately formats system messages for guard rejections by using the correct reason property (event.reason), improving clarity and consistency in agent responses.

[v0.245.6] - 2025-10-01 - Containerization of Aihub Bot for Enhanced Deployment

Added

✨ Introduced Dockerfile for Aihub Bot: Added a comprehensive Docker configuration to containerize the Aihub Bot, enabling consistent and portable deployments across various environments.
🚀 Streamlined Build and Runtime Environments: Implemented a multi-stage Docker build process to optimize image size and efficiency, ensuring a lean runtime environment.
🔒 Enhanced Operational Security and Dependency Management: Configured a non-root user for improved security and integrated Poetry for robust Python dependency management within the container.

[v0.245.5] - 2025-09-30 - Infrastructure and Developer Experience Enhancements

Added

⚙️ Introduced set-latest GitHub Action: A new automated workflow for managing container image tags, ensuring the latest tag is correctly applied to published images, improving deployment consistency.

Refactor

🧹 Refined Model Data Fetching: Updated the web application's SDK to use clearer getModels API calls and ModelTypeGroupDtoReadable types, enhancing code readability and maintainability.
⚡️ Optimized ESLint Configuration: Streamlined the ESLint setup by expanding global ignore rules to skip unnecessary files and simplifying plugin imports, leading to faster linting and a cleaner development environment.

Refactor

🧹 Standardized Event Types: Aligned internal naming conventions by updating the ChunkEventReadable type to ChunkEvent within the event component resolution logic, enhancing code consistency.

Changed

🔄 Updated LLM Event Logging: Adjusted the LLMEvent to correctly capture the chat model name using the model_name property from the agent configuration, ensuring more accurate and consistent event metadata.

Refactor

🧹 Improved .gitignore: Added a rule to ignore GitHub Copilot related files, streamlining the development environment setup and reducing repository clutter.

[v0.245.2] - 2025-09-24 - Introducing Advanced Guardrails & Granular Event Visibility

Added

🔑 Sensitive Information Guard: Introduced a new guard mechanism to detect and redact sensitive or confidential information from LLM responses, significantly enhancing data privacy and security.
🦾 Granular Guard Events: Implemented a comprehensive suite of new guard events, including AgentSuitabilityAcceptEvent, AgentSuitabilityRejectEvent, ContextSufficientAcceptEvent, ContextInsufficientRejectEvent, FewShotAcceptEvent, FewShotRejectEvent, SensitiveInfoAcceptEvent, and SensitiveInfoRejectEvent. These events provide detailed insights into guard decisions at various stages of agent execution.
📄 Language Detection Event: Introduced LanguageEvent to explicitly capture and signal the detected language of a user request, enabling more robust internationalization within agent workflows.
🖼️ Raw Event Data Viewer: Added the capability to view the raw JSON data for any event directly within the frontend, improving debugging and transparency for developers.
🧪 Expanded Frontend Testing Agent: The FrontendTestingAgent now includes new steps to demonstrate various guard mechanisms and common workflow events, making it a powerful tool for UI development and testing.
🚀 New LLM Model Support: Added support for the upcoming gpt-5 series of models and minimal reasoning effort to the API models.
⚙️ Enhanced Tool Calling Flexibility: Introduced more flexible tool calling options, including custom tool definitions and explicit allowed tool choices for richer agent-LLM interactions.
⚡️ LLM Stream Obfuscation Option: Added an include_obfuscation option to LLM stream requests, allowing for more control over sensitive information during streaming.
🌐 Comprehensive i18n for Guards and Events: Extended internationalization support to cover all new guard events and significantly improved descriptions for various common events in multiple languages.

Changed

🔄 Standardized Agent Guard Integration: The FewShotAgent and RAGAgent have been updated to utilize the new standardized AgentSuitabilityAcceptEvent, AgentSuitabilityRejectEvent, and other specific guard events, ensuring consistent guardrail application across agents.
🖼️ Improved Source Node Display: RAG agent's source node display in the frontend has been significantly improved, providing clearer grouping by document, chunk counts, and an explicit message when no relevant chunks are found.
💬 Enhanced Chat History Event Display: The LimitChatHistoryEvent now offers a more detailed and user-friendly display on the frontend, showing the limited messages clearly.
❓ Refined Standalone Question Display: The StandaloneQuestionCondenserEvent now has a dedicated and improved display in the UI, making the condensed question more prominent.
🔍 Clearer Reranker Output: The RerankerEvent display has been updated to explicitly highlight the top_k relevant documents chosen by the reranker.

Fixed

🐛 API Endpoint Duplication: Addressed a minor issue in the API configuration where a NotificationController endpoint was inadvertently registered twice.

Refactor

🧹 Unified Guard Result Structure: Introduced a base GuardResult model and refactored existing guards (agent_description_guard, context_sufficient_guard, few_shot_guard) to inherit from it, standardizing the output and improving maintainability.
⚡️ Decoupled Guard Rejection: The GuardRejectionEvent now correctly inherits from GuardEvent instead of StopEvent, providing a clearer separation of concerns between guard decisions and workflow control.
🗑️ Internal Event Cleanup: Removed deprecated RightAgentEvent and consolidated internal agent-specific guard event classes into the shared aihub_lib.
⚙️ Refined Async Test Utility: The async_test decorator in the testing utilities now correctly returns the result of the decorated asynchronous function, improving testing flexibility.

Added

✨ Anonymous File Access Storage Configuration: Introduced a new environment variable, ANONYMOUS_FILE_ACCESS_SERVICE_STORAGE_BACKEND, to configure the storage backend for anonymous file access services.
🦾 LLM Message Preprocessing: Implemented automatic preprocessing for LLM messages with a new PreprocessingOpenAILike wrapper, enhancing model compatibility and performance by merging consecutive messages of the same role.
📄 Summary Node Content Type Metadata: Added NODE_CONTENT_TYPE metadata to summary nodes generated by the RecursiveSummaryParser, providing richer context and improving document structure representation.
✨ Gemma 3 Multimodal Small Model Support: Integrated comprehensive configuration for the local/gemma-3-multimodal-small model within LiteLLM, enabling its usage across various agents and pipelines.
🚀 Llama-CPP Cache Volume: Introduced a dedicated cache volume (llamacpp-cache) for the llama-cpp service to optimize performance and improve stability of local LLM inference.
🔑 HuggingFace API Key for LiteLLM: Added HUGGINGFACE_API_KEY to LiteLLM environment variables, facilitating seamless integration with HuggingFace models.
🛠️ Docling Model Permission Helper: Included a new docling-model-permission service to ensure correct file permissions for Docling models, improving container startup reliability.

Changed

🔄 Default LLM Model Update: Switched the default Large Language Model from local/qwen3-small to local/gemma-3-multimodal-small across LLMWrappingAgent, LlamaIndexAgent, RAGAgent, WebUI, and the default RAG pipeline to leverage improved capabilities.
⚡️ RAG Agent Retrieval Configuration: Optimized retrieval parameters in the RAG Agent by setting check_context_sufficiency to False and reducing retrieve_k and retrieve_prev_next.num_nodes from 20/10 to 5, streamlining retrieval efficiency.
⚙️ Pipeline Configuration Simplification: Streamlined S3 data lake and Milvus vector store configurations within the playground pipeline by removing redundant directory_name and namespace_name parameters.
🚀 Updated Llama-CPP Service and Configuration: Upgraded the llama-cpp Docker image and updated its command-line arguments to officially support the gemma-3-4b-it model and enable Flash Attention (--flash-attn on) for enhanced inference speed.
⬇️ Pinned LiteLLM Service Version: Explicitly pinned the LiteLLM service to version v1.50.0 for improved stability and compatibility across the system.

Fixed

🐛 Robust NATS Event Serialization: Improved the serialization logic for NATS BaseEvent to ensure proper handling of nested ChatMessage and BaseModel instances, preventing potential serialization errors and enhancing data integrity.

Refactor

🧹 RAG Agent Test Setup: Refactored the RAG Agent testing fixtures by introducing a dedicated test_collection setup and teardown, which improves test isolation and reliability for vector store operations.
🔄 Pipeline Job Parameter Renaming: Updated pipeline job definitions to use source_location_name instead of namespace_name, aligning with recent Dagster API changes for better clarity and consistency.

[v0.245.0] - 2025-09-18 - Comprehensive Distributed Tracing with OpenTelemetry

Added

✨ Introduced Comprehensive Distributed Tracing: Integrated OpenTelemetry across all AI-Hub services to provide end-to-end visibility into complex AI workflows.
🔭 New OpenTelemetry Collector Service: Added a dedicated OpenTelemetry Collector to the development environment, simplifying observability data ingestion and routing to various backends like Phoenix.
⚙️ Centralized OpenTelemetry Configuration: Implemented a new OpenTelemetrySettings module for unified management of tracing and logging providers across all Python services.
🏷️ Explicit Function Tracing: Introduced the @trace_fn decorator for explicit tracing of key service and persistence methods, automatically capturing their inputs and outputs.
🚫 Selective Tracing Control: Added the @no_trace decorator and SmartTracer to enable selective disabling of OpenTelemetry instrumentation for specific functions or modules, optimizing trace data for relevance.
📚 Deep Observability Documentation: Introduced new documentation for "Deep Observability with OpenTelemetry" and an Architectural Decision Record (ADR) detailing the rationale and implementation of end-to-end tracing.
📦 Automated Library Instrumentation: Implemented AihubInstrumentor for automatic instrumentation of critical libraries like pymongo, milvus, redis, requests, httpx, aiohttp, jinja2, botocore, llama_index, and logging, ensuring broad coverage without manual code changes.
📊 LLM Client Trace Propagation: Enhanced LLM client configurations to automatically propagate OpenTelemetry trace context to underlying model calls, enabling full traceability of generative AI interactions.

Changed

🔄 Updated NATS Client API for Observability: Modified NATS Publisher and Subscriber constructors to require explicit naming, which improves trace readability and component identification in observability tools.
🚀 Optimized NATS Development Configuration: Adjusted NATS development server settings (e.g., max_payload, max_connections) for better performance and resource utilization in local environments.
🗺️ Restructured Documentation Paths: Updated documentation paths for feature overviews and internal command documentation, enhancing overall clarity and consistency.
📊 Enabled OpenTelemetry for LiteLLM: Configured LiteLLM to activate its native OpenTelemetry callback and telemetry, extending distributed tracing to all LLM router interactions.
🧩 Enhanced FastAPI OpenTelemetry Integration: Implemented FastAPI-specific OpenTelemetry instrumentation, including automatic span enrichment with user, service, and request context for API endpoints.

Refactor

🧹 Overhauled Agent Tracing Architecture: Replaced the legacy RunTraceCoordinator with a new AgentRunTracer for more robust and flexible agent run and step tracing using a two-span approach.
📂 Consolidated OpenTelemetry Infrastructure: Centralized OpenTelemetry-related utilities and base instrumentation within aihub_lib/infrastructure/opentelemetry for improved modularity and maintainability.
🔗 Streamlined Context Initialization: Refactored RunContext and ThreadContext initialization in agents to use a new for_topic class method, simplifying context management from NATS topic information.

Removed

🗑️ Deprecated Legacy Tracing Modules: Removed RunTraceCoordinator and the generic tracing decorator, which have been superseded by the new OpenTelemetry instrumentation framework.

Added

✨ Introduced ImageLoader: A new dedicated document loader for image files, designed to wrap images in HTML figure tags for pipeline ingestion, enabling specialized processing without performing OCR or detailed image analysis.
🚀 Integrated ImageLoader into Document Parsing: The new ImageLoader is now utilized within the DocumentParserResource to specifically handle a comprehensive range of image formats (jpg, jpeg, png, gif, bmp, tiff, webp, tif, heif).

Removed

🗑️ Deprecated Asynchronous Loading in Document Intelligence Loader: The aload_data asynchronous method has been removed from the DocumentIntelligenceLoader, simplifying its API surface.
🗑️ Image Support from Document Intelligence Settings: Direct support for image file types (jpg, jpeg, png, bmp, tiff, heif) has been removed from AzureDocumentIntelligenceSettings, aligning the Document Intelligence loader's focus solely on structured document processing.
🗑️ Image Support from Docling Settings: Image-related file types (image, jpg, jpeg, png, tif, tiff, bmp, webp) have been removed from DoclingSettings, specializing the Docling loader for textual and specific document formats.

Refactor

🔄 Centralized Image File Handling: The overall strategy for processing image files has been refactored, dedicating the new ImageLoader to all image formats and removing redundant image handling capabilities from DocumentIntelligenceLoader and DoclingLoader. This improves modularity and clarifies responsibility across document loaders.

Refactor

🧹 Namespace Entity field renamed: Renamed the last_updated field to updated_at in the NamespaceEntity for improved consistency and clarity in data lake persistence.

[v0.244.0] - 2025-09-15 - Knowledge Base Evolution: Streamlined Uploads, Localized Collections, and Improved Pipeline Harmony

Added

🚀 Direct File Uploads API: Introduced new API endpoints for initiating and validating direct file uploads to the datalake, and for retrieving a list of supported file types.
🖼️ Frontend Document Upload Modal: Implemented a new user interface component allowing drag-and-drop file uploads directly into knowledge collections.
✨ Knowledge Collection Management API: New API endpoints for creating and updating knowledge namespaces (now referred to as collections).
🎨 Frontend Collection Management UI: Added user interfaces for creating new knowledge collections and editing existing ones, including display names and descriptions.
🗄️ Datalake Entity Models: Introduced BucketEntity and NamespaceEntity for a more structured and persistent representation of datalake containers and their collections.
🗣️ LLM-Powered Translation Service: A new TranslationService to automatically localize collection display names and descriptions using an integrated Large Language Model.
📄 Unified Document DTO: Added DocumentDTO to consistently represent both fully ingested (processed) and in-progress (processing) documents across the API and UI.
🛡️ File Type Configuration: Implemented FileTypeConfig to centralize the management of supported file extensions and their corresponding MIME types, enhancing upload validation.
⚡️ Knowledge Service Caching: Integrated ttl_cache for efficient caching of knowledge service data, improving performance for database and document listings.
🌐 Expanded Internationalization: Added comprehensive French and Italian translations for core API prompts and all new user interface elements, alongside new i18n keys for better clarity.
🛠️ Pipeline Bucket Utilities: Introduced new utility functions (bucket_utils) to dynamically manage the mapping between datalake buckets/directories and knowledge base entities.
📦 Container-Level Data Lake Resource: Added DataLakeContainerResource to enable pipeline operations across entire datalake containers, supporting multi-namespace processing.
🤖 Agent HITL Events: Enhanced AgentDTO with hitl_request_events and hitl_response_events to support Human-in-the-Loop workflows for agents.

Changed

📝 Enhanced Knowledge Collection Listings: The knowledge base now visually differentiates and lists both successfully ingested documents and those still undergoing processing.
📊 Enriched Knowledge Base Metadata: Knowledge databases (buckets) and collections (namespaces) now feature localized display names, descriptions, and an auto_sync status, providing more context in the UI.
🔄 API Route Standardization: Renamed /file API routes to /files for improved consistency across the API surface.
⚙️ Refined Pipeline Operations: RAG pipelines are now optimized to dynamically resolve namespaces and operate on a container-wide basis, adapting to the new structured datalake model.
💡 Updated Default LLM Configurations: Adjusted the default embedding and language model configurations within the RAG pipeline to better suit new requirements.
🌍 Locale Handling Extension: Updated LocaleHandler to seamlessly process LocaleStringEntity objects, improving i18n integration with persistent data.
🗃️ Nullable Document Content: The content field in IngestedDocument is now nullable, accommodating scenarios where document content may not always be present or fully parsed.

Removed

🗑️ Legacy Activity Tracking: Eliminated deprecated endpoints and associated DTOs related to historical spending and activity tracking.
🧹 Obsolete gpt-5 Model References: Removed gpt-5 model names from the ChatCompletionRequest enum, streamlining model selection options.
✂️ Simplified Tool Call Types: Removed support for custom tool calls and simplified related types in the ChatCompletion API.
🚫 Redundant Event Fields: Removed reasoning_content from ChunkEvent and ThoughtEvent and include_obfuscation from ChatCompletionStreamOptionsParam.
⚠️ minimal Reasoning Effort: The minimal option for reasoning effort in ChatCompletionRequest has been removed.
❌ Old Namespace Model: The legacy Namespace Pydantic model has been replaced by the new NamespaceEntity and NamespaceDTO.
⛔ txt from Docling Extensions: Removed .txt from the Docling supported extensions list, deferring to the new FileTypeConfig for document type management.
🗺️ Directory-Specific Data Lake Arguments: Removed the directory_name argument from various Data Lake resources and operations, shifting to a more flexible, container-based processing model.
🗑️ DocStore Namespace Configuration: The namespace_name field has been removed from DocStoreResource, centralizing namespace management within the new datalake entities.

Refactor

🦾 Standardized Datalake Access: Abstract AnonymousFileAccessService now includes generic methods for generating upload URLs, verifying file existence, and listing files, with concrete implementations for Azure and S3 services.
🔀 Centralized Knowledge Listing Logic: The get_databases logic in KnowledgeService has been thoroughly refactored to leverage the new BucketEntity and NamespaceEntity data models, providing a unified source of truth.
🌐 Optimized i18n Translation Workflow: Reorganized the i18n translation structure to seamlessly integrate with the new LLM-based translation service.
⚙️ Streamlined Pipeline Resources: Refactored pipeline asset factories and jobs to align with the new, simplified data lake resource model, reducing redundancy and improving maintainability.

Removed

🗑️ Removed the extensive SDK Architecture documentation from the Quick Start section, which previously detailed the aihub_lib, aihub_agent, aihub_pipeline, aihub_process, aihub_api, and aihub_bot packages, as well as cross-package integration patterns. This content, previously marked as Work In Progress (WIP), has been temporarily cleared.
🗑️ Removed the in-depth "Why our SDK" documentation from the Quick Start guides, which provided a comprehensive comparison with other AI frameworks and platforms. This content, previously marked as Work In Progress (WIP), has been temporarily cleared.

[v0.243.9] - 2025-09-11 - Fine-Grained Control Over Document Image Extraction

Added

⚙️ include_images Configuration: Introduced a new configurable option include_images in the DocumentParserResource, enabling users to specify whether images should be extracted and embedded during document parsing.
🖼️ Image Inclusion Parameter: Added include_images as a new parameter to both DoclingLoader and DocumentIntelligenceLoader, providing explicit control over image extraction at the individual loader level.
📄 Figure Content Management: Implemented the remove_figure_tags_keep_content utility within DocumentIntelligenceLoader to ensure textual content from figures is retained even when image embedding is disabled.

Changed

🔄 Conditional Image Processing: Updated DoclingLoader and DocumentIntelligenceLoader to honor the new include_images setting, allowing for more flexible document processing where images can be optionally excluded from the final document content.
⚡️ Pipeline Integration: The parse_document_from_data_lake operation now passes the include_images setting from the DocumentParserResource to the document loaders, ensuring consistent behavior across the pipeline.

[v0.243.8] - 2025-09-11 - Enhanced Asynchronous Document Loading and Docling Integration

Added

⚡️ Asynchronous Document Processing: Introduced asynchronous aload_data methods to both DoclingLoader and DocumentIntelligenceLoader, enabling non-blocking document ingestion and improved concurrency.
🚀 Asynchronous Docling API Interaction: Implemented a comprehensive asynchronous client for the Docling API, including robust polling mechanisms for monitoring and retrieving results from long-running jobs.
⚙️ Docling Polling Configuration: Added new configuration settings, POLL_INTERVAL and MAX_POLLS, to allow fine-tuning the behavior of asynchronous Docling API polling.
🧹 Internal Processing Helpers: Introduced _process_docling_response and _build_request_body helper methods to encapsulate common logic and improve code clarity within the document loaders.

Changed

🔄 Async Data Lake Parsing: The parse_document_from_data_lake operation in the data lake pipeline now leverages the new asynchronous document loading capabilities, improving overall pipeline efficiency.
📄 Docling Output Streamlining: The DoclingLoader now primarily requests json output from the Docling service, with Markdown content generated internally from the structured JSON for a more streamlined and flexible process.
🖼️ Improved Docling Image Handling: Enhanced the internal mechanism for exporting and embedding images into Markdown content generated by the DoclingLoader, delegating image-to-markdown conversion to the Picture object itself.

Refactor

🦾 Consolidated Loader Logic: Extracted repetitive document processing steps into dedicated, internal helper methods across DoclingLoader and DocumentIntelligenceLoader, reducing code duplication and enhancing maintainability.

[v0.243.7] - 2025-09-11 - Streamlined Release Tagging Workflow

Refactor

🔄 Refactored Release Tagging Workflow: Improved the internal GitHub Actions workflow responsible for creating and pushing release tags. This update ensures that any last-minute changes are incorporated into the main release commit via an atomic amend operation, and utilizes a safer force-with-lease push strategy. Additionally, redundant tag verification steps have been removed to streamline the overall release process.

[v0.243.6] - 2025-09-10 - Robust Release Tagging Workflow

Fixed

🐛 Corrected Tagging Reference: Ensured that release tags consistently point to the correct merge commit or the latest main branch tip in GitHub Actions, particularly for squash and rebase merges, preventing tags from being created against incorrect SHAs.

Changed

🚀 Improved Tag Verification Process: Enhanced the reliability and feedback of the post-push tag verification in GitHub Actions by directly fetching remote tags and performing a local check, providing clearer success and failure indications during the release tagging workflow.

[v0.243.5] - 2025-09-09 - Enhanced CI/CD Tag Verification Reliability

Changed

⚡️ Improved Tag Verification Reliability: The GitHub Actions workflow for adding release tags now includes multiple retry attempts with a brief delay, making the remote tag verification process more robust and resilient to transient network issues or eventual consistency delays.

[v0.243.4] - 2025-09-09 - Documentation & Local Setup Enhancements

Added

🏠 Enhanced Local Development Setup: Introduced comprehensive documentation for running the AI-Hub locally, including details on SSL support, custom domain routing, and configuration with docker-compose.local.yml for a more production-like development environment.
📄 Expert-Agent Feature Overview: Added a new documentation page providing an overview of the Expert-Agent capabilities within the SDK.

Changed

🌐 Documentation Navigation Improvements: Updated numerous internal links throughout the SDK documentation from absolute to relative paths, enhancing navigation robustness and portability.

Added

✨ New Documentation Structure & Content: Introduced a completely reorganized documentation site, moving from a previous three-part structure to a comprehensive five-part journey: Vision & Positioning, Platform Journey, SDK Journey, Ecosystem Journey, and Reference & Resources. This includes entirely new sections on the problem solved, our solution, the Swiss approach, and detailed architecture.
🎨 Interactive Documentation UI: Enhanced the documentation site with new Vue components for interactive navigation boxes and dynamic gradient backgrounds, significantly improving user experience and visual appeal.
🚀 Refined Core Platform Deployment: Introduced docker-compose-core.dev.yml, a comprehensive Docker Compose configuration that integrates all core infrastructure services for robust local development and streamlined production deployments.
📦 SDK Quickstart Examples: Added new Python files (my_document_pipeline.py, simple_pipeline.py) to the aihub_pipeline/playground/quick_start/ directory, providing runnable examples for building data ingestion pipelines.
⚡ Pipeline Quickstart Tooling: Added a quickstart target to the aihub_pipeline Makefile for easily running quickstart pipeline examples.
🌐 Frontend Development Variables: Added WEBUI_URL and WS_ENDPOINT to .env.dev to streamline local frontend development.
🧹 Temporary File Exclusion: Added .tmp* to .gitignore in aihub_pipeline to ignore temporary build files.

Changed

⚙️ Deep OpenWebUI Integration & Configuration: Expanded the open-webui service configuration with extensive environment variables, enabling granular control over authentication, storage, model access, PII handling, web search, image generation, code interpretation, audio features, and user permissions.
📖 Documentation Navigation & Styling: Updated documentation front matter indices and integrated new custom CSS for enhanced theming, fonts, scrollbars, and improved Mermaid diagram rendering across the site.
🔗 Documentation Internal Linking: Updated internal links within aihub_bot/README.md, aihub_lib/README.md, aihub_agent/README.md, aihub_api/README.md, and aihub_process/README.md to align with the new documentation structure.
📄 Unified Root README Location: Modified the sync-docs.sh script to correctly place the root README.md within the new documentation structure at docs/6_code_deep_dive/1_introduction/index.md.
🚀 Docling UI Enabled by Default: Enabled the Docling service UI in docker-compose.dev.yml for improved visibility during development.
🦶 Updated Footer Copyright: Changed the footer message in the documentation site.

Removed

🗑️ Obsolete Documentation Content: Deleted outdated documentation sections related to 1_business, 2_developers, and 3_features, which have been superseded by the new, comprehensive documentation structure.
🖼️ Deprecated Architecture Diagrams: Removed old architecture diagrams (tier1.png, tier2.png, etc.) as they have been replaced by new, detailed diagrams reflecting the updated platform vision.

[v0.243.2] - 2025-09-09 - Streamlined Release Process and Robust Tagging

Changed

✅ Enhanced Initial Tagging Robustness: Implemented pre-checks to prevent duplicate tags and added explicit error handling during the initial tag creation and push, ensuring greater reliability.
🚀 Improved Final Release Tagging: Upgraded the final tag creation to use annotated tags and incorporated a robust verification step to confirm successful remote tag updates.
🔐 Scoped Git Configuration: Adjusted Git user configuration within workflow steps to use local scope, enhancing security and preventing unintended global state changes.

Refactor

🧹 Streamlined Release Workflow: Consolidated various dependency management, version updating, and commit operations from a standalone job directly into the release finalization process, simplifying the overall workflow.
🔄 Optimized Job Dependencies: Removed an intermediate update-and-dependencies job, streamlining the flow between tag computation, changelog generation, license reporting, and final release actions.

[v0.243.1] - 2025-09-03 - Advanced RAG Agent, Secure Local Dev & OpenWebUI Enhancements

Added

🦾 Advanced RAG Agent: Introduced a new Retrieval Augmented Generation (RAG) agent, designed for sophisticated knowledge retrieval and integration into workflows.
🚀 Secure Local Development Setup: Implemented a comprehensive local development environment (docker-compose.local.yml) featuring Traefik, self-signed SSL certificates, and custom nip.io domains for a more realistic and secure local testing experience.
🔑 make local-cert Command: Added a convenient Makefile target to generate local SSL certificates effortlessly using mkcert for localhost and nip.io domains.
📄 Expanded Local Development Documentation: Provided extensive new documentation in the README.md to guide users through setting up the enhanced local AI-Hub stack with HTTPS.
⚡️ OpenAI-Compatible OpenWebUI Pipeline: Introduced a new OpenWebUI pipeline (openai_pipeline.py) that offers a standard OpenAI-compatible interface for accessing AI-Hub models, complementing the existing event-based agent integration.
🔔 Notification System Foundation: Added a new NotificationController to the API, laying the groundwork for future in-application notifications.
⚙️ Dagster Local Workspace Configuration: Included a workspace.local.yaml to simplify the setup and management of Dagster pipelines in local development environments.

Changed

🔄 Core Library Dependency Management: Transitioned aihub_lib dependencies across all core projects (agent, API, bot, pipeline, process) from local paths to specific Git tags, ensuring more stable and versioned dependency resolution.
🛡️ OpenWebUI API Key Security: Enhanced security for OpenWebUI's RAG, Audio STT/TTS, and Image Generation API calls by switching to the SUPERUSER_TOKEN.
🌐 Configurable Phoenix Endpoint: Updated the Phoenix client to utilize a configurable endpoint, offering greater flexibility for integration with monitoring and evaluation systems.
📊 Refined LiteLLM Guardrail Defaults: Adjusted LiteLLM guardrail defaults for Presidio (mask and block) to be off by default and narrowed the default presidio_language to "de" for optimized control.
🧹 Standardized Milvus Bucket Naming: Aligned and standardized the default Milvus bucket name in MinIO configurations to "milvus" across environments.
📦 Updated Docling Service: Upgraded the Docling service to v1.3.1 and refined its Docker Compose configuration for improved model loading and permission handling.
🚀 OpenWebUI Version Update: Upgraded the OpenWebUI image to v0.6.22, incorporating the latest features and bug fixes.
📝 Streamlined AI-Hub API OpenAI Endpoints: Simplified the API's OpenAI endpoints by consolidating methods like get_models_with_assistants to get_models, removing redundant "with assistants" naming.
🔒 Enhanced User Header Encoding: Implemented base64 encoding for the X-OpenWebUI-User-Name header in OpenWebUI interactions, improving security and compatibility.

Removed

🗑️ Deprecated Pipeline Runner Settings File: Removed PipelineRunnerSettings.py to streamline configuration, with settings now primarily managed through environment variables.

[v0.243.0] - 2025-09-02 - Refined Image Tagging for Build Action

Changed

⚡️ Improved Image Tagging: The build_image GitHub Action has been updated to use comma-separated values for secondary image tags, ensuring more robust and widely compatible tag assignment for published images.

[v0.242.0] - 2025-09-02 - New Model Catalog and Advanced AI Features

Added

🚀 Introduced a comprehensive Models overview: Users can now browse and view detailed information about all available AI models, including their capabilities, pricing, and rate limits, through a new dedicated section in the application.
🦾 New Model Information API: A dedicated API endpoint now provides structured access to retrieve lists of AI models, grouped by type, and detailed information for individual models.
📈 Core Spending and Activity Tracking DTOs: Added new data transfer objects to support detailed tracking and breakdown of model usage, costs, and API activity.

Changed

🤖 Expanded Chat Completion Models and Parameters: The chat completion API now supports new gpt-5 series models and introduces a minimal option for reasoning effort, offering more flexibility in AI interactions.
💬 Enhanced Event and Message Schemas: Updated core event and message structures (ChunkEvent, ThoughtEvent, UserMessageEvent) to include additional context such as reasoning_content and model_name, improving traceability and debugging.
⬆️ Dependency Updates: Upgraded various internal and external dependencies to their latest versions, improving performance, stability, and security across the platform.
🌐 Internationalization for Model Features: Added new translations for German, English, French, and Italian to support the new model browsing experience.

Removed

🗑️ Streamlined Agent DTOs: Removed hitl_request_events and hitl_response_events fields from the AgentDto schema, simplifying agent configuration related to human-in-the-loop interactions.

[v0.241.2] - 2025-08-29 - Streamlined Builds, Dependencies, and Naming

Added

🧰 Introduced format-md-win Makefile target: Provides a dedicated command for formatting Markdown files specifically for Windows development environments, enhancing cross-platform tooling.

Changed

🚀 Optimized Docker image builds: Production Docker images for agents, APIs, and pipelines now install dependencies without development packages, significantly reducing image size and build times.
📦 Promoted tomlkit to core dependency: The tomlkit library is now a primary project dependency, reflecting its expanded use beyond development workflows.

Refactor

🔄 Standardized changelog file naming: Renamed changelog.md to CHANGELOG.md across the repository for better consistency and adherence to common project conventions, affecting GitHub Actions, documentation scripts, and changelog generation.
🧹 Streamlined mdformat-vuepress dependency: The mdformat-vuepress plugin is now managed as a standard package rather than a local path dependency, simplifying project setup and maintenance.

Removed

🗑️ Deprecated local mdformat-vuepress plugin files: Removed the now-obsolete local mdformat-vuepress source files as the plugin is integrated via a standard package manager.

[v0.241.1] - 2025-08-28 - Enhanced Docker Image Tagging and Build Control

Added

✨ New secondary_tag input: Introduced an optional secondary_tag input to all Docker image build workflows, enabling more flexible tagging for images (e.g., latest, nightly).

Changed

🔄 Conditional Docker Tagging Logic: The image build action now intelligently applies secondary Docker tags only when a value is explicitly provided, preventing the creation of unwanted or empty tags.
⚙️ Dynamic Secondary Tag Assignment: Build workflows have been updated to dynamically assign secondary Docker tags based on the trigger event, defaulting to nightly for repository_dispatch events, or utilizing the user-specified secondary_tag input.

[v0.241.0] - 2025-08-28 - Live Thinking Agents and Seamless Open-WebUI Streaming

Added

✨ Introduced Native Open-WebUI Integration: A brand-new aihub_integration module enables seamless, event-driven communication with Open-WebUI via Server-Side Events (SSE), providing a richer and more interactive user experience.
⚡️ Enabled Live Agent Thoughts: The core streaming mechanism now parses <think> tags within LLM outputs, allowing agents to stream their internal reasoning ("thoughts") in real-time alongside regular content.
🦾 Exposed Human-in-the-Loop (HITL) Events: Agents can now declare and discover HumanInTheLoopRequestEvent and HumanInTheLoopResponseEvent types, enabling interactive workflows where human input is explicitly requested.
📄 Dedicated SSE Streaming API Endpoints: New API endpoints /api/v1/agents/{class}/{id}/{event}/stream are introduced to support Server-Sent Events, facilitating real-time updates and interactive agent experiences for external UIs like Open-WebUI.
📚 Integrated Docling for Document Processing: Switched to a new DoclingLoader for robust document parsing, enhancing the platform's ability to ingest and process various document formats.
🔑 Configurable S3 URL Signing Secret: Added S3_STORAGE_URL_SIGNING_SECRET environment variable for enhanced security when generating pre-signed URLs for S3 storage.
⚙️ New Docling Service Initialization Script: A dedicated entrypoint script for the Docling service now automates model downloading and ensures proper setup with correct permissions.
📊 Added aihub_integration to Core Makefile Checks: Ensured linting, formatting, and type-checking for the new aihub_integration module.
🚀 New Open-WebUI Actions for Tracing and Sources: Pipeline actions source_action.py and tracing_action.py are now integrated into the aihub_integration module, providing direct access to agent execution traces and document sources from within Open-WebUI.

Changed

🔄 Refactored Streaming Logic for Thought Events: The EventDisplayer now uses a new StreamProcessor and TagParser to intelligently handle and display both regular and "thinking" content during LLM streaming.
⬆️ Updated RAG Agent Configuration: The default RAG agent now uses local/qwen3-small and an improved embedding model (azure/text-embedding-3-large), with enhanced retrieval settings (retrieve_k to 20, node_types to include "summary") and a comprehensive system prompt.
🚚 Unified Docker Volume Paths: Standardized local development/deployment volume paths from local-volumes to .docker-volumes across all Docker Compose configurations, improving consistency.
📦 Upgraded Third-Party Docker Images: Updated Open-WebUI to v0.6.22, Docling to v1.3.1, and llama-cpp to server-cuda-b5490 with updated model configurations.
🔌 Enhanced Open-WebUI Connection Parameters: Added specific AI-Hub connection details (AIHUB_BASE_URL, AIHUB_SUPERUSER_API_KEY, etc.) and updated database/S3 endpoints in Open-WebUI Docker configurations for better integration.
🛡️ Improved Authentication Strategy Order: The TokenAndOauth2Handler now prioritizes bearer token authentication before OAuth2, ensuring more flexible and robust authentication flows.
⏳ Increased Default LLM Request Timeout: The default timeout for LLM API requests has been extended from 60 to 600 seconds to accommodate longer processing times.
🗑️ Simplified Chat Message Structure: AssistantChatMessage and UserChatMessage are deprecated and replaced by the more generic ChatMessage from llama_index.core.base.llms.types, simplifying message handling across the platform.
🌐 Improved Agent Discovery DTOs: Agent DTOs (AgentClassDTO, AgentDTO, AgentInstanceDTO) now include hitl_request_events and hitl_response_events, providing more complete information about agent capabilities.
🎨 Revised WebUI Styling: Updated chat message background colors and event display headers for a more polished user interface.
🛠️ Refined Docling API Integration: The DoclingLoader now interacts with the Docling service using a revised API endpoint and request body format for more reliable document conversions.
📝 Updated Event Display Components: Frontend components for displaying events (ChunkEvent, ThoughtEvent, RetrieverEvent, etc.) have been updated to align with the latest event schema changes and improved user experience.
🔍 Optimized Logging for Libraries: Default logging level for third-party libraries is now set to WARNING during development/testing to reduce log verbosity.
♻️ API Path Regex Syntax Fixed: Corrected regex patterns for path and query parameters in API routes for consistency and proper validation.

Removed

🗑️ Deprecated Open-WebUI OpenAI Proxy Pipelines: The legacy aihub_connection.py and aihub_connection_with_file_sending.py Open-WebUI pipelines, which mimicked the OpenAI API, have been removed in favor of the new, native SSE integration.
🧹 Removed Redundant Chat Message Types: Custom AssistantChatMessage and UserChatMessage definitions were removed, simplifying the messaging interface by using the universal ChatMessage type.
🗑️ Consolidated LiteLLM Model Configurations: The local/qwen3-0.6b and local/text-embedding-gte models have been removed from LiteLLM configurations, streamlining available model options.
🗑️ Removed Reasoning Content from ChunkEvent: The reasoning_content field has been removed from ChunkEvent, now being explicitly handled by the ThoughtEvent.

[v0.240.6] - 2025-08-26 - S3 Security Hardening and Configuration Alignment

Security

🔑 Enhanced Secret Key Handling: Updated S3 resource configurations to securely access secret keys using .get_secret_value(), reinforcing protection for sensitive credentials across S3 file access services and data lake clients.

Changed

🔄 Standardized S3 Endpoint Configuration: Renamed the S3 configuration parameter ENDPOINT_URL to ENDPOINT for consistency across all related S3 services and resources, simplifying configuration management.

Changed

🚀 Upgraded aihub_pipeline Docker images to Python 3.13 from 3.11, leveraging the latest language features and performance improvements for pipeline execution.
⚙️ Enhanced CI/CD workflows to support private GitHub dependencies by configuring SSH, facilitating more robust and secure access to internal repositories during automated builds.
📦 Streamlined Docker build processes by temporarily including the mdformat-vuepress module in several service images (aihub_agent, aihub_api, aihub_pipeline) to ensure internal tool compatibility.

[v0.240.4] - 2025-08-15 - Enhanced CI/CD Actions with Python Version Flexibility

Changed

⚡️ Configurable Python Version for Backend Actions: The lint_backend and test_backend GitHub Actions now support a new python_version input. This allows users to specify the Python version used for linting and testing within their workflows, providing greater flexibility while maintaining 3.13 as the default.

[v0.240.3] - 2025-08-15 - Core Infrastructure Enhancements and Frontend Streamlining

Added

⚙️ New Infrastructure Components: Introduced Traefik as an ingress controller and OAuth2 Proxy for authentication, enhancing deployment capabilities and security.
🚀 Expanded Dagster Deployment Options: Added new dagster:latest and dagster:nightly Docker images, providing more flexible options for pipeline orchestration.
🔒 SSH Support for GitHub Actions: Enabled SSH authentication within GitHub Actions to ensure more secure and reliable Git operations during release workflows.
🐍 Python 3.13 Support in CI: Updated GitHub Actions to utilize Python 3.13, aligning with the latest Python version for improved compatibility and future-proofing.

Changed

🔄 Updated Playwright Version: Upgraded the Playwright Docker image to v1.54.1-jammy, incorporating the latest browser automation features and stability improvements.
⚡️ Streamlined CI/CD Dependency Management: Refactored Poetry installation and dependency management steps across GitHub Actions workflows for enhanced efficiency and consistency.

[v0.240.2] - 2025-08-14 - Architectural Evolution: Containerized Deployment and New AI Capabilities

Added

🦾 Containerized Deployment Architecture: Introduced comprehensive Docker Compose files (docker-compose.latest.yml, docker-compose.nightly.yml, docker-compose-gpu.*.yml) to enable multi-environment deployments for the entire AI-Hub platform, supporting local development, nightly builds, and production-ready setups.
📄 Architectural Decision Records: Added new documentation (arc42/decisions) detailing the rationale and design of the new containerized deployment and microservice build pipeline architectures.
⚙️ Modular Dockerfiles for Services: Introduced dedicated Dockerfiles for api, bot, dagster, web, llm_wrapping_agent, and default_rag_pipeline to enable component-specific image builds and deployments.
✨ LLMWrappingAgent: A new generic agent capable of wrapping and exposing Large Language Models as a service within workflows, simplifying LLM integration.
⚡️ DefaultRAGPipeline: A new default Retrieval-Augmented Generation (RAG) pipeline, providing out-of-the-box capabilities for knowledge retrieval and content generation.
🔑 Milvus Configuration Settings: Introduced a new MilvusSettings class to standardize Milvus database connection URL and embedding dimension configuration across the platform.
📦 Multi-Environment Configuration Files: Added environment-specific configuration files for Dagster, LiteLLM, Milvus, and NATS, allowing tailored setups for different deployment environments.

Changed

🔄 CI/CD Release Workflow Redesign: Reworked the GitHub Actions release pipeline (add-tag.yml) to support a more granular, event-driven build and release process for individual microservices.
🚀 Docker Image Build Action Enhancements: Updated the shared build_image action to support secondary tags (e.g., nightly) and passing version as a build argument, improving flexibility and versioning.
📈 Agent Test Stability: Increased delay_before_stop in agent test runs to enhance stability and reliability during asynchronous test execution.
🔒 Web UI Container Security: Hardened the Web UI Docker image by introducing a dedicated non-root user for improved security posture.
🧹 Markdown Formatting Enforcement: Integrated mdformat into PR readiness, changelog, and license check steps to ensure consistent Markdown formatting across the repository.
🔗 Milvus Integration Update: Updated API and pipeline configurations to leverage the new standardized MilvusSettings, ensuring consistent and configurable Milvus connectivity.

Fixed

🐛 API Dockerfile Path Fix: Corrected a minor syntax error in the API Dockerfile's PATH environment variable, ensuring proper script execution.

Refactor

📐 Core Application Restructuring: Restructured the main application entry points for aihub_api and aihub_bot into app/main.py files, improving modularity and maintainability.

[v0.240.1] - 2025-08-13 - Enhanced Documentation Formatting and Tooling

Added

✨ Introduced Standardized Markdown Formatting: Added mdformat and various plugins to enforce consistent and high-quality markdown formatting across all documentation files.
⚙️ Custom VuePress Markdown Support: Developed and integrated a new mdformat-vuepress plugin to ensure proper formatting and preservation of custom VuePress ::: block syntax in documentation.
🚀 New format-md Command: Introduced a make format-md command to easily apply the new markdown formatting standards across the project.

Refactor

🧹 Unified Documentation Styling: Applied a comprehensive set of markdown formatting rules across all documentation files, improving readability and consistency.

[v0.240.0] - 2025-08-12 - Streamlined Setup with Superuser Authentication and Enhanced Security

Added

✨ Global Superuser Authentication: Introduced a new superuser authentication system, enabling immediate API access for external services in containerized or backend-only deployments via a configurable environment variable token. This streamlines initial setup and simplifies service integration.
🦾 Automatic Default Role Initialization: Added the capability to automatically create standard AI-Hub roles (e.g., AIHubUser, AIHubAdmin) during application startup, simplifying initial deployment and configuration steps.
📄 Comprehensive Authentication Settings: Introduced new environment variables and settings for general API access control, OpenWebUI signing secrets, and configurable OAuth identity provider selection, offering more granular control over authentication strategies.

Changed

🔄 Dynamic Authentication Handler Initialization: Refactored the core authentication handler (TokenAndOauth2Handler) to dynamically configure available authentication strategies (superuser, token, OAuth2, OpenWebUI) based on environment settings, significantly streamlining setup and enabling more flexible deployments.
⚡️ Enhanced OpenWebUI Authentication: Improved the OpenWebUI authentication mechanism by migrating from a custom hashing scheme to a more robust HMAC-SHA256 signature verification, bolstering security against request tampering.
📝 ADR Documentation Improvements: Updated the Architecture Decision Records (ADR) process documentation with clearer guidance on focusing decisions and a simplified date format for new ADR files.

Security

🔑 Robust Secret Management: Migrated the handling of all sensitive configuration parameters (e.g., API keys, connection strings, private tokens) to Pydantic's SecretStr type. This prevents accidental exposure of secrets in logs, debugging output, or system representations, significantly enhancing application security.

Refactor

🧹 Standardized Application Runner API: Renamed the method for obtaining the application instance from get_app() to create_app() across all core runners (ApiRunner, BotRunner, and the base Runner), improving API consistency and clarity.
🔄 Generalized Identity Provider Naming: Renamed MultiStrategyTokenIdentityProvider to MultiStrategyIdentityProvider to accurately reflect its role in aggregating various identity provider types, not just token-based ones.

[v0.239.0] - 2025-08-08 - Architectural Overhaul: Streamlined AI Integration and Enhanced Development Workflow

Added

✨ Unified Infrastructure Configuration: Introduced a new configs/ directory with litellm/config.dev.yaml, milvus/milvus.dev.yaml, minio/minio-entrypoint.sh, and postgres/init-multiple-dbs.sh, providing centralized and detailed configurations for LiteLLM, Milvus, MinIO, NATS, and PostgreSQL, enabling more flexible and powerful development environments.
🐳 New Docker Compose Files for Development: Added docker-compose.dev.yml (CPU) and docker-compose-gpu.dev.yml (GPU) for simplified local infrastructure setup, including services like LiteLLM, llama.cpp, HuggingFace TEI, and improved health checks.
📦 Enhanced Frontend Build and Deployment: Introduced a multi-stage Dockerfile, nginx.conf, config.template.json, and plugins/config-loader.client.ts for aihub_web, enabling production-ready Nginx-based serving with runtime environment variable injection.
⚙️ Windows Image Mirroring Utility: Added mirror-image-win target to aihub_iac/Makefile for convenient Docker image mirroring on Windows environments.
🧪 Standardized Local Docker Run Configurations: New .idea/runConfigurations XML files were added to streamline building specific Docker images (API, Bot, LLM Wrapping, RAG Pipeline) and running infrastructure on CPU/GPU directly from IntelliJ/PyCharm.
📄 New Development Environment File: Introduced a .env.dev file to standardize local development environment variables, including API keys and service endpoints.

Changed

🚀 Optimized CI/CD Backend Testing: Refined GitHub Actions workflows for backend testing by updating runner types to ubuntu-latest-8-cores, switching to direct Docker Compose service specification, enabling comprehensive caching for Poetry and HuggingFace models, and implementing pre-pulling of Docker images for faster test runs.
⬆️ Updated Python Requirement: The core project now officially supports and recommends Python 3.13, and related documentation in README.md and aihub_doc has been updated.
⚡️ Increased Test Stability: Extended the timeout for Docker service health checks in CI/CD workflows to accommodate varying startup times and enhance test reliability.
🔒 Flexible Image Build Authentication: The SSH key requirement for the aihub_action/build_image workflow has been made optional, providing more flexibility in build environments.
🔄 Refined Makefile Commands: Added run-prod targets for aihub_api and aihub_bot for standardized Gunicorn-based production server startup, and introduced use-local-core-without-install/use-remote-core-without-install for more granular dependency switching.
🌐 Web Frontend Application Structure: Migrated the Nuxt.js frontend from a generic .playground to a dedicated .app directory, centralizing application-specific configurations and build artifacts.
📐 Milvus Vector Store Dimension: Increased the default embedding vector dimension for Milvus in agent and pipeline configurations from 768 to 1024 to support higher-dimension embedding models.

Removed

🗑️ Legacy Configuration Files: Eliminated numerous outdated .env files from individual microservice playgrounds, centralizing environment variable management to the root .env.dev for consistency.
🧹 Deprecated IntelliJ/PyCharm Run Configurations: Removed old Docker run configurations (Infra_CPU.xml, Infra_GPU.xml, Webui.xml) that have been superseded by the new, more flexible configurations.

Refactor

🏗️ Comprehensive Settings System Overhaul: Replaced the fragmented *Config.py and *Access.py classes (e.g., ApiConfig, CosmosAccess, NatsConfig, PhoenixConfig, RedisConfig, S3Config, DoclingConfig) with a new, unified EnvironmentSettings hierarchy. This centralizes environment variable loading and makes configuration more consistent and maintainable across the entire platform.
🧠 Unified AI Model Integration with LiteLLM: Performed a significant architectural shift by refactoring all LLM, Embedding, Image Generation, Speech-to-Text, and Text-to-Speech model configurations (ChatLLMConfig, EmbeddingLLMConfig, AzureOpenAILLMConfig, SelfHostedEmbeddingConfig, AzureImageModelConfig, etc.) into a streamlined system (LLMConfig, EmbeddingModelConfig) that leverages LiteLLM as a central proxy. This greatly simplifies model management, enables dynamic model routing, and reduces hardcoded dependencies on specific AI providers within the codebase.
🧹 Streamlined Frontend Scripts and Paths: Updated aihub_web/package.json scripts and .gitignore to reflect the new .app directory structure, ensuring a cleaner and more organized frontend project.
🧽 Cleaned Up aihub_lib Structure: Removed various redundant or now obsolete model and infrastructure configuration files and their corresponding __init__.py files, significantly reducing complexity and improving code clarity in the core library.
⚙️ Standardized Auth Settings Usage: Migrated authentication logic across agents and API to consistently use the new DangerousDevelopmentOnlyAuthSettings and OAuth2Settings, ensuring uniform handling of fake users and OAuth2 configurations.
📦 Streamlined Poetry Dependency Management: Modified aihub_agent, aihub_api, aihub_bot, aihub_pipeline, and aihub_process pyproject.toml files to use local path dependencies for aihub_lib, simplifying local development and CI/CD builds.
🧪 Optimized Agent Test Setup: Improved agent test runners to copy .env.dev to .env for consistent environment loading during tests and streamlined Poetry dependency installation in CI without redundant poetry lock calls when using local cores.
💬 Improved LLM Cost Reporting: Updated EventDisplayer to use the model_name attribute from the new LLMConfig for more accurate and consistent cost reporting.

[v0.238.1] - 2025-08-07 - Empowering Developers: New Documentation, Enhanced CI/CD, and AI Assistant Integration

Added

📚 Comprehensive Documentation System: Introduced a completely new, structured documentation site using VitePress, replacing the old markdown files and setting up automated deployment to GitHub Pages. This includes a new dynamic sidebar and custom theming.
🧑‍💻 AI Assistant Development Commands: Added a suite of new commands for Claude Code (.claude/commands/), including create-pr, document-decisions, document-feature, explain, implement-feedback-from-pr, and update-doc, to streamline developer workflows.
🔗 Model Context Protocol (MCP) Integration: Implemented MCP support within the AI-Hub, enabling AI coding assistants to observe and interact with running services (API, Phoenix, MongoDB) for enhanced debugging and development.
🚦 Automated Version Label Enforcement: Introduced a new CI check to enforce version bump labels (major, minor, patch) on all pull requests targeting the main branch, ensuring release predictability.
⚙️ PR Agent Configuration Customization: Added a new .pr_agent.toml file for comprehensive customization of the PR agent's behavior, offering granular control over automated code reviews.
🤖 Standardized AI Context Files: New CLAUDE.md and GEMINI.md files were added across scopes to provide context for AI coding assistants.
✅ Prohibited Force Pushes: Added Git settings to prohibit force pushes to the main branch, enhancing codebase integrity.
🧪 Consolidated Local Testing: Added explicit test targets to Makefiles in core microservices (aihub_agent, aihub_api, aihub_bot, aihub_lib, aihub_process) for easier and consistent local test execution.

Changed

🚀 Streamlined CI/CD Versioning: The automated release workflow now triggers only on merged pull requests to main and dynamically determines the version bump (major, minor, patch) based on PR labels, providing more granular control over releases.
🧠 Updated AI Code Review Model: The PR review workflow now utilizes the o4-mini AI model and updated API versions (2024-12-01-preview), enhancing performance and compatibility for automated code reviews.
📈 Improved CI/CD Robustness: Configured various GitHub Actions workflows to use fail-fast: false, allowing other jobs to continue execution even if one job fails, improving overall CI stability.
🔄 Refined ApiRunner and BotRunner: Restructured runners to consistently utilize Starlette as the base application, simplifying internal architecture and supporting multiple mounted applications like the new MCP server.
📦 Standardized Monorepo Dependencies: Shifted aihub_lib dependencies in microservices to local path references (develop = true) instead of Git tags, streamlining local development and cross-project dependency resolution.
🔑 Simplified API Playground Authentication: The default authentication in the aihub_api playground has been changed to DangerousDevelopmentOnlyAuthHandler for easier local development setup.
🐍 Updated Python Action Examples: Modified Python version in aihub_action/README.md examples from 3.13 to 3.11 for relevant actions.

Removed

🗑️ Deprecated Documentation System: Eliminated the entire old documentation system (aihub_doc/ markdown files and arc42/ structure), replaced by the new VitePress-based system.
🧹 Obsolete gitkeep Files: Cleaned up empty .gitkeep files across the repository.
✂️ Redundant PR Agent Hardcoded Settings: Removed hardcoded PR agent auto-review settings from the GitHub Action, now managed via .pr_agent.toml.

Refactor

📊 Unified Python Testing: Standardized test targets in Makefiles across all Python microservices, simplifying local test execution.
🎯 Streamlined CI/CD Configuration: Centralized PR agent settings into a dedicated .pr_agent.toml file, improving maintainability of automated review processes.
🏗️ Refined API and Bot Runner Structures: Simplified base application creation and mounting logic within ApiRunner and BotRunner for a cleaner architecture.

[v0.238.0] - 2025-08-06 - Internal Version Alignment

Changed

⬆️ Internal Version Alignment: Updated the version numbers across all core microservices (aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, aihub_pipeline, aihub_process) and synchronized internal dependency tags to v0.238.0, ensuring consistent versioning for the new release.

[v0.237.0] - 2025-08-06 - GitHub Action Streamlining

Removed

🗑️ Removed GitHub Release creation: The automated step within the build image action that previously created a GitHub Release for the built Docker image has been removed.

[v0.236.0] - 2025-08-06 - RAGAgent: Enhanced Control with System Prompts

Added

✨ System Prompt Configuration for RAGAgent: Introduced the capability to configure the RAGAgent with a system prompt. This allows for more precise guidance of the underlying Large Language Model's behavior and responses, with built-in support for multi-language prompts using LocaleString.
🧪 Comprehensive System Prompt Testing: Added new test scenarios to validate the RAGAgent's multi-language system prompt functionality, ensuring accurate prompt application and response generation.

Changed

🔗 Streamlined Monorepo Dependency Resolution: Python microservices (aihub_agent, aihub_api, aihub_bot, aihub_pipeline, aihub_process) now reference the aihub_lib core library using Git tags instead of local paths, simplifying CI/CD and enhancing production deployment consistency. This change affects how Python dependencies are reported in the LICENSES.md file.
⚡️ Extended RAGAgent Test Timeout: Increased the default delay before stopping RAGAgent test runs to provide more robustness and accommodate longer LLM response times.

Refactor

🧹 Optimized Test Fixture Scope: Changed the self_hosted_agent_config pytest fixture to be session-scoped, improving test execution performance by setting up the configuration once per test session instead of per function.

[v0.235.0] - 2025-08-06 - Platform Upgrade and Enhanced License Compliance

Added

✨ Automated License Compliance Reporting: Introduced a new, comprehensive system to automatically scan, categorize, and report on all Python, Node.js, and Docker image dependencies, generating a detailed LICENSES.md file for improved transparency and compliance.
🔊 Python 3.13 Audio Compatibility: Added the audioop-lts dependency to the API service to ensure seamless audio processing compatibility with Python 3.13.

Changed

🚀 Python 3.13 Support: Upgraded all core Python projects to officially support and leverage Python 3.13, enabling access to the latest language features and performance improvements.
⚙️ Updated CI/CD Workflows: Enhanced GitHub Actions workflows for tagging, linting, and testing to align with Python 3.13 and integrate the new license checking processes, improving build reliability and compliance checks on pull requests.
📄 Documentation Updates: Reflected the new Python 3.13 requirement in the main README.md and action documentation for clearer setup instructions.

Refactor

🧹 Type Hint Modernization: Migrated typing_extensions.override annotations to Python's built-in typing.override for cleaner and more consistent type declarations, aligning with Python 3.13 standards.

[v0.234.2] - 2025-08-07 - Enhanced PR Agent Configuration and Performance Updates

Added

✨ Introduced pr_agent.toml configuration file: Enables comprehensive customization of the PR agent's behavior, including advanced settings for auto-approval, reasoning effort, code suggestions, and best practices, empowering users with greater control over automated reviews.

Changed

🚀 Updated AI Model and API Versions: Upgraded the review-pr workflow to utilize the o4-mini AI model and the 2024-12-01-preview API version, enhancing performance and ensuring compatibility with the latest service capabilities.

Refactor

🧹 Centralized PR Agent Settings: Migrated several hardcoded PR agent configurations from action.yml into the new pr_agent.toml file, streamlining configuration management and improving overall maintainability.

[v0.234.1] - 2025-08-07 - Platform Evolution: Python 3.13, Automated Compliance, and RAG Agent Enhancements

Added

✨ Automated License Compliance Reporting: Introduced a comprehensive system to scan, categorize, and report on all Python, Node.js, and Docker image dependencies, generating a detailed LICENSES.md file for improved transparency and compliance.
🔊 Python 3.13 Audio Compatibility: Added the audioop-lts dependency to the API service to ensure seamless audio processing compatibility with Python 3.13.
🦾 RAGAgent System Prompt Configuration: Introduced the capability to configure the RAGAgent with a system prompt, allowing for more precise guidance of the underlying Large Language Model's behavior and responses. This includes support for multi-language prompts and comprehensive testing.
🛡️ Enforced Main Branch Protection: Implemented Git settings to explicitly prohibit force pushes to the main branch, enhancing codebase integrity and preventing accidental history rewrites.
✅ Automated PR Version Label Check: Introduced a new CI check that ensures all pull requests targeting main have a major, minor, or patch label, streamlining version management and release predictability.

Changed

⬆️ Python 3.13 Platform Upgrade: All core Python projects and CI/CD workflows (aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, aihub_pipeline, aihub_process) have been upgraded to officially support and leverage Python 3.13, enabling access to the latest language features and performance improvements.
⚙️ Refined CI/CD Versioning Strategy: The automated release workflow now triggers only on merged pull requests to main and determines the version bump (major, minor, or patch) based on PR labels, providing more granular control over releases.
⚡️ Increased RAGAgent Test Timeout: Extended the default timeout for RAGAgent test runs to provide more stability and accommodate longer LLM response times in complex scenarios.
🔗 Internal Dependency Tag Synchronization: Synchronized all internal aihub_lib dependency tags across microservices to align with the latest versioning, ensuring consistent module integration.
🧪 IntelliJ/PyCharm Test Runner Updates: Updated various IntelliJ/PyCharm run configurations for Python tests, aligning them with pytest and modern Poetry virtual environment practices.
📝 Documentation Python Version: Updated README.md to reflect the new Python 3.13 requirement.
🌐 Web Project Name: Renamed the aihub_web project in package.json for consistency.

Removed

🗑️ Deprecated GitHub Release Creation: Eliminated the automated step within the build image GitHub Action that previously created a GitHub Release for each built Docker image, streamlining CI/CD processes.

Refactor

🧹 Python 3.13 Type Hint Modernization: Migrated typing_extensions.override annotations to Python's built-in typing.override and simplified AsyncGenerator type hints for cleaner and more consistent type declarations, aligning with Python 3.13 standards.
🧪 Optimized Agent Test Fixture Scope: Changed the self_hosted_agent_config pytest fixture to be session-scoped, improving test execution performance by setting up the configuration once per test session instead of per function.
📦 Streamlined Poetry Setup in CI: Refined the Poetry installation and caching steps in GitHub Actions workflows, improving CI efficiency and consistency.

[v0.234.0] - 2025-08-04 - Major Notification System Overhaul and Web Experience Improvements

Added

🚀 Comprehensive User Notification System: Introduced a robust notification system to keep users informed with real-time updates and an intuitive interface for managing alerts. This major addition includes:
- ✨ New Notification API Endpoints: Established a full suite of API endpoints for retrieving paginated notifications, updating individual notification statuses, and performing bulk actions (e.g., marking multiple as read or done).
- 🔔 Dedicated Web User Interface: Launched a new "Notifications" page on the web application, offering a centralized view with advanced filtering options (unread, to-do, all), pagination, and bulk management capabilities.
- 🚨 Real-time Notification Overlay and Poller: Integrated a dynamic notification poller and an interactive overlay component (a bell icon with a counter badge) directly into the top navigation bar. New notifications are now highlighted via toast messages for immediate awareness and quick access.
- 📚 Persistent Notification Data Model: Developed a new database entity and associated service logic for efficient storage and management of all user notifications.
🌍 Expanded Internationalization for RAG Agent: Added German, French, and Italian translations for the RAG Agent's name and description, broadening language support for core functionalities.
✅ Enhanced Web Application Readiness Checks: Incorporated pnpm lint into the aihub_web Makefile's pr-ready target, strengthening automated code quality and consistency checks for web changes.

Changed

💬 Improved API Error Messaging: Enhanced the web client's error handling to provide more detailed and user-friendly error messages in toast notifications, pulling detail information directly from API responses.
🎨 Refined Popover Component Styling: Updated the visual styling of Popover components across the web UI to ensure a consistent and modern aesthetic.
🌐 Improved User Experience and Accessibility for Top Bar: Added tooltips and enhanced aria-label attributes to the dark mode toggle and notifications icon in the top navigation bar, improving user guidance and accessibility.

Refactor

🔄 Standardized Agent Event Topic Naming: Renamed the AgentTopic schema and its associated types in the web SDK to AgentInstanceTopic for clearer terminology and improved consistency within the event topic structure.

Fixed

📄 Corrected Localization for Process Title: Addressed an issue where the "Processes" heading on the user detail page was not correctly localized, ensuring proper display across all supported languages.

Removed

🗑️ Deprecated User Avatar Component: Removed the standalone User/Avatar.vue component, as its functionality has been streamlined and integrated directly into the User/Bar.vue component for better modularity.

[v0.233.0] - 2025-07-30 - Internal Dependency Enhancements

Changed

⚡️ Internal Dependency Updates: Upgraded core libraries to enhance stability and prepare for future improvements.

[v0.232.0] - 2025-07-29 - Enhanced Dashboard Trend Clarity

Changed

💡 Adjusted Trend Indicator Logic: The trending up/down visual indicators for Exception Events on dashboards now correctly reflect their severity, displaying green for trending down (good) and red for trending up (bad) to provide more intuitive visual feedback.

[v0.231.0] - 2025-07-29 - Enhanced Agent Testing with Custom Events

Added

🚀 Customizable Agent Simulation Events: Introduced the ability to specify custom start_events and stop_events for the SimulatedAgentApiTestRunner, greatly expanding the flexibility for testing agent behavior and API interactions under diverse conditions.
🧪 New Test Event Definitions: Added TestStartEvent and TestStopEvent classes to facilitate the creation of unique testing scenarios for agent event processing.
✨ Comprehensive Custom Event Test Suite: Implemented a new test suite (test_agent_api_with_custom_event.py) to thoroughly validate the handling of custom start and stop events within the agent API, ensuring robust and predictable agent responses.

Changed

⬆️ Version Updates: Bumped project versions across all core components (aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, aihub_pipeline, aihub_process) and their internal aihub_lib dependencies to v0.231.0, ensuring consistent dependency alignment.

Removed

🗑️ Internal Claude AI Integration Scripts: Removed development-specific scripts related to Claude AI integration, streamlining the project's internal tooling.

[v0.230.0] - 2025-07-29 - NATS Topic Management and Refined Process Configurations

Added

✨ Introduced WalkthroughContext for Processes: A new Redis-backed context specifically designed for process walkthroughs, enabling robust and persistent state management across complex, multi-step processes.
📦 New Process Configuration Persistence Model: Implemented dedicated persistence entities (ProcessConfigEntity, ProcessConfigEntityDocument, ProcessConfigEntityEmbeddedDocument) to support flexible storage and retrieval of process configurations. This includes both class-level defaults and instance-specific overrides stored in the database.
🚀 ProcessConfigSpecs for Discovery: Incorporated ProcessConfigSpecs into process discovery events, allowing clients to programmatically inspect the schema and parameters required for process configurations, enhancing dynamic integration.
💡 Direct Event Creation Utilities: Added from_raw_data class methods to StartEvent and ProcessStartEvent, simplifying the creation of these events from raw input data within API and runner contexts.
✅ Comprehensive Process Dispatcher Tests: Introduced a new, extensive test suite for ProcessDispatcher to ensure reliable event handling, configuration management, and step execution within processes.
🧪 Enhanced Process Service Database Integration Tests: Added new tests for ProcessService and ProcessConfig database operations, validating correct interactions with the new persistence model.
🤖 API Endpoint for Agent StartEvent: Exposed a new API endpoint enabling the direct sending of StartEvent to agents, which facilitates programmatically initiating agent runs with specified configurations.
🧪 New Test Fixtures: Added cleanup_db_and_cache fixtures for agent and process services to improve test isolation and ensure clean state in integration and unit tests.

Changed

📈 Refined Agent and Process API Discovery: Updated the API discovery mechanisms in AgentController and AgentService to simplify agent listing, and revamped process discovery in ProcessService to provide more detailed and cache-aware retrieval of process instances and classes.
📄 Event Deserialization Flexibility: Modified DispatchableWorkflow to allow its get_steps_waiting_for_event method to accept WorkEvent types, broadening the dispatcher's capability to react to various process-related events.
🗄️ Improved Event Persistence Logic: Updated PersistedAgentEventEntity and PersistedProcessEventEntity to correctly utilize the new AgentInstanceTopic and ProcessInstanceTopic respectively, ensuring accurate data storage with the refined topic structure.
🧩 Extended WorkRequestEvent with process_id: Added a process_id field to WorkRequestEvent to enhance clarity and provide a direct association with the relevant process instance.
✍️ Enhanced Logging for Subscribers: Improved logging for JSSubscriber to provide more detailed information during subscription lifecycle events (start and stop).
⚙️ Process Runner Configuration: ProcessRunner now accepts default_process_config and dynamically handles process configuration loading, distinguishing between class and instance-specific settings.

Refactor

🔄 NATS Topic Hierarchy Refinement: Restructured NATS topics for Agents and Processes to introduce clear distinctions between class-level and instance-level communication. This included renaming existing topics (e.g., AgentTopic to AgentInstanceTopic, ProcessTopic to ProcessInstanceTopic) and introducing new ones (AgentClassTopic, ProcessClassTopic, PartialProcessTopic). Related TopicManager and Subscriber classes were updated to align with this new hierarchy.
🧹 Streamlined Process Discovery Events: Consolidated and refined process discovery events, replacing a monolithic ProcessDiscoveryResponseEvent with ProcessClassDiscoveryResponseEvent and ProcessInstanceDiscoveryResponseEvent for more granular information exchange during discovery.
⚡️ Improved Dispatcher stop Method: Enhanced BaseDispatcher's stop method to ensure more robust cleanup of event stores upon dispatcher termination, improving resource management.
⚙️ Centralized Process Configuration Handling: Modified ProcessDispatcher to robustly manage process configurations by prioritizing explicit configurations provided in ProcessStartEvent and gracefully falling back to default process configurations, leveraging the new WalkthroughContext.
🏗️ Core Context Relocation: Relocated the foundational BaseContext class from aihub_agent to aihub_lib, promoting its reusability across all core service modules.
🗑️ Delegator Topic Manager Alignment: Refactored Agent and Process delegators to align their topic management with the new ProcessClassTopicManager and ProcessInstanceTopicManager, improving consistency and maintainability.
🧹 Agent & Process Service Cleanup: Streamlined internal methods in AgentService and ProcessService by adding _ prefix to denote private helper methods and cleaning up redundant caching calls in tests.
🛠️ Process Entity Persistence: Updated ProcessEntity to use ProcessConfigEntityDocument for referenced configurations and ProcessConfigEntityEmbeddedDocument for default configurations, enhancing process config persistence.

Removed

🗑️ Deprecated Generic Discovery Events: Eliminated the generic DiscoveryRequestEvent and the monolithic ProcessTopic class. These have been superseded by more specific and granular event and topic types, improving type safety and clarity in discovery and messaging workflows.
🧹 Redundant Agent Config Start Event Test: Removed a test file (test_agent_config_start_event.py) as its functionality is now fully covered by updated and more comprehensive dispatcher tests.

[v0.229.0] - 2025-07-25 - Multi-Cloud Storage for Files and Pipelines

Added

☁️ Multi-Cloud Storage Abstraction: Introduced a foundational AbstractAnonymousFileAccessService and concrete implementations for Azure Blob Storage and S3/MinIO, enabling consistent file access across various cloud providers.
⚙️ Configurable File Access Service: Added FileAccessServiceConfig to dynamically configure and instantiate the appropriate file access service based on environment settings, simplifying multi-cloud deployment.
🛠️ S3/MinIO Data Lake Integration: Implemented S3DataLakeIOManager and supporting S3DataLakeClient and S3DataLakeFileSystem resources for Dagster pipelines, bringing robust S3-compatible storage capabilities.
🏗️ Abstracted Data Lake Clients: Introduced base interfaces (AbstractDataLakeClient, AbstractDataLakeClientResource, AbstractDataLakeFileSystemResource) to standardize interactions with different data lake storage solutions.
⚙️ S3 Configuration: Added a new S3Config for centralizing MinIO/S3 connection parameters and credentials.

Changed

🔄 Unified File URL Generation: Updated aihub_api and aihub_lib to leverage the newly introduced multi-cloud file access services for generating secure, temporary file URLs.
🖼️ Enhanced Image Resolution: Improved the combine_nodes_in_order utility in aihub_lib and the frontend image component to correctly resolve and display images from both Azure and S3/MinIO storage.
🚀 Cloud-Agnostic Pipeline Operations: Migrated core pipeline operations (e.g., fetching, deleting files) and their associated resources to utilize the new abstract data lake client, improving flexibility and maintainability.
🧪 Playground Configuration: Adjusted playground examples to showcase the new S3/MinIO data lake capabilities and switched the default document parser to DOCUMENT_INTELLIGENCE.
⬆️ Dependency Updates: Upgraded Dagster and related dependencies, and added boto3 and s3fs to support comprehensive S3/MinIO integrations.

Removed

🗑️ Azure AI Search Vector Store: Discontinued support for Azure AI Search as a vector store option within aihub_pipeline, streamlining the available vector store integrations.
🗑️ Deprecated DataLakeFile.from_uri Static Method: Removed the static method for creating DataLakeFile from a URI, as its logic is now encapsulated within cloud-specific DataLakeClient implementations.

Refactor

🧹 Infrastructure Configuration Restructure: Moved NatsConfig and RedisConfig into dedicated infrastructure subpackages for clearer logical grouping and improved project structure.
📄 Improved Docstring Clarity: Cleaned up and removed redundant parameter descriptions from various Infrastructure-as-Code (IAC) and API test fixture docstrings.
⚙️ Centralized DataLakeFile Creation: Refactored the instantiation logic for DataLakeFile objects by moving the URI-based creation method into cloud-specific data lake client implementations for better encapsulation and maintainability.

[v0.228.0] - 2025-07-25 - Azure Document Intelligence Key Access Alignment

Refactor

🔄 Updated Azure Document Intelligence Key Access: Aligned the method for retrieving primary keys from Azure Document Intelligence services with recent API changes, ensuring robust and consistent authentication.

[v0.227.0] - 2025-07-23 - Improved Azure Identity Service Robustness

Fixed

🖼️ Azure User Profile Image Fetching: Enhanced the retrieval of user profile images from the Azure Graph API to be more resilient to various API errors. The service now gracefully handles non-404 errors by logging warnings and consistently returns None when an image cannot be fetched, preventing unexpected exceptions and improving stability.

Refactor

🧹 Import Statement Placement: Relocated the asyncio import to the top-level of the AzureGraphService module, aligning with best practices for improved code readability and organization.

Refactor

🧹 Consolidated Data Transfer Objects (DTOs): Renamed AgentClass to AgentClassDTO and introduced AgentInstanceDTO, relocating all agent-related DTOs into a dedicated dto subdirectory for improved organization and clarity.
🔄 Enhanced DTO Encapsulation: AgentInstanceDTO now manages its own persistence by handling entity creation/updates and generating discovery response events directly, improving data model autonomy.
🗄️ Decoupled Agent Persistence: The AgentEntity creation and update logic was refactored to accept explicit parameters instead of DTO objects, enhancing the reusability and flexibility of the persistence layer.
⚙️ Standardized Event Specification Handling: Updated EventSpec creation methods to consistently use EventSpecs objects, improving internal data handling.
🔗 Streamlined Topic Inheritance: AgentInstanceDiscoveryTopic now correctly inherits from AgentClassDiscoveryTopic, reducing redundancy and clarifying topic structure.

Changed

📄 Agent Configuration Data Format: The ChatService now passes agent configuration data as a dictionary instead of a Pydantic object when publishing UserMessageEvent, standardizing data exchange.

Removed

🗑️ Deprecated Agent DTO Files: Removed the old AgentInstance.py and aihub_api/aihub_api/agents/__init__.py files as part of the DTO consolidation and relocation effort.
🧹 Redundant NATS Topic Managers: Cleaned up unused and overlapping topic management methods within AgentClassTopicManager for a leaner API.
🚫 Transferred Discovery Event Logic: Removed the from_agent_instance method from AgentInstanceDiscoveryResponseEvent as its functionality has been moved to the AgentInstanceDTO for better encapsulation.

[v0.225.0] - 2025-07-21 - Enhanced Document Structure and Content Rendering

Added

✨ Hierarchical Document Headings: Implemented the dynamic rendering of document headings (H1-H6) directly within the context. This provides a clearer hierarchical structure and significantly improves readability and navigation for complex documents by showing heading changes as content progresses.
📄 Structured Content and Summary Tags: Individual node content is now explicitly wrapped in XML-like tags (<content> or <summary>) based on its type. This provides more precise semantic context and better structural clarity for each block of text.
🔒 HTML Escaping for Content: Enhanced the integrity and security of rendered content by automatically escaping HTML special characters in both headings and content. This prevents unintended rendering issues and potential injection vulnerabilities in the generated output.

Changed

⚙️ Updated Default Document Language: The default language for newly ingested nodes has been updated from English to German, better aligning with regional configurations.
⚡️ Refined Node Sorting Logic: Improved the internal sorting mechanism for nodes within a document. Nodes are now ordered more accurately based on their section start lines and type, which is crucial for the correct presentation of the new hierarchical structure.
🧹 Streamlined Timestamp Handling: An internal utility for formatting Unix timestamps has been removed, as date and time handling for document metadata is now managed more efficiently by the underlying system.

[v0.224.0] - 2025-07-21 - Flexible Agent Configuration and Advanced Discovery

Added

✨ Dynamic Agent Configuration: Agent configurations can now be dynamically provided via a StartEvent or loaded from a database, offering greater flexibility and allowing custom overrides of default agent settings.
🚀 Enhanced Agent Discovery Mechanism: Implemented granular discovery for agent classes and specific agent instances, enabling the system to distinguish between an agent's base definition and its runnable, configured versions.
🗄️ Database Persistence for Agent Configurations: Custom agent configurations can now be saved and retrieved from the database, enabling persistent and shareable agent setups across deployments.
🔌 New Pydantic Models for Vector Store Configurations: Introduced dedicated Pydantic configurations for AzureAISearchVectorStore and MilvusVectorStore (AzureAISearchVectorStoreConfig, MilvusVectorStoreConfig), allowing for direct embedding and validation of vector store settings within agent configs.
📈 agent_class Field in AgentConfig: A new agent_class field has been added to AgentConfig for consistent agent class identification throughout the system.
📄 save_config.py Example: A new example demonstrating how to persist agent configurations to the database for custom or production deployments.
🧪 Comprehensive Tests for Agent Config/Discovery: Added extensive new tests covering the new dynamic configuration and discovery mechanisms within the AgentDispatcher and AgentService.
🌐 API Acts as a Discoverable Agent Instance: The API now actively participates in the instance-level agent discovery, allowing it to expose configured agent instances dynamically.
🌍 LocaleStringEntity for Database Localization: Introduced a dedicated persistence entity for localized strings, enabling more robust internationalization within the database.

Changed

⚙️ Refined Agent Configuration Loading Logic: The agent dispatcher now intelligently loads configurations, prioritizing those provided in a StartEvent, then database-persisted configurations, and finally falling back to the default_agent_config defined in the runner.
🏗️ Updated Agent Runner Initialization: Agent runners now explicitly accept a default_agent_config parameter, clearly defining the base configuration for new agent runs.
💡 Improved Type Specificity for LLM and Embedding Configs: Agent configurations (e.g., in RAGAgent) now utilize more precise type unions for LLMs and embedding models, enhancing type safety and clarity.
🔗 Streamlined Vector Store Integration: RAG and Retrieval agents now leverage the .to_llama_index() method directly from the new Pydantic vector store configurations for more consistent and encapsulated integration.
⚡️ Enhanced Background Task Management: Improved handling of asyncio.Task instances within dispatchers, subscribers, and tracers to ensure proper lifecycle management and prevent premature garbage collection.
🌐 Revised API Agent Discovery: The API now discovers and exposes full agent instances (including their specific configurations) instead of just class definitions, reflecting the new dynamic configuration capabilities.
📚 Documentation Updates: Agent examples in aihub_doc/5_agents_in_detail.md revised to reflect updated AgentConfig and removed system_prompt.
🏷️ Agent Class Naming Consistency: Agent class names are now consistently PascalCase in API tests for improved standardization.

Fixed

🐛 Robust Profile Image Fetching: Improved AzureGraphService to gracefully handle 404 responses when fetching user profile images, preventing errors for users without an image.
📝 Typo in Process Description: Corrected a minor typo in the ProcessDTOSchema description within the web SDK.

Refactor

🔄 Consolidated Agent Configuration Management: Standardized agent configuration handling across the system, enabling dynamic overrides and database persistence for agent instances.
✨ Refined Agent and Event Discovery Architecture: Overhauled the discovery mechanism to support both class-level and instance-level agent introspection, providing more granular control and information.
🏗️ Reorganized API DTOs for Agent Representation: Introduced dedicated DTOs (AgentClass, AgentInstance, MinimalAgentDTO) to clearly separate the concept of an agent's definition from its configured, runnable instances in the API.
📦 Streamlined Vector Store Configuration: Replaced direct factory calls for MilvusVectorStore and AzureAISearchVectorStore with new Pydantic-based *VectorStoreConfig models and their to_llama_index() method, simplifying configuration.
🧹 Renamed Discovery Topic Types: Renamed DISCOVERY_TOPIC to INSTANCE_DISCOVERY_TOPIC and added CLASS_DISCOVERY_TOPIC for clearer distinction in NATS topics.
🧹 API Endpoint Path Naming: Renamed _get_endpoint_name to _get_endpoint_base_path for clarity in API endpoint construction.
📁 Project Structure Alignment: Reorganized test file directories under aihub_api/playground/testing/tests for better logical grouping.

Removed

🗑️ Deprecated Configuration Fields: The system_prompt, color, and voice fields have been removed from the base AgentConfig and AgentConfigDTO, streamlining the core configuration and allowing for more flexible, agent-specific prompt and UI customization to be defined in agent subclasses.

[v0.223.0] - 2025-07-18 - Streamlined Pipeline Operations and Enhanced Robustness

Added

🦾 New Job Creation Utility: Introduced materialize_asset_job to simplify the creation of jobs for materializing specific selections of assets.
📅 Automated Playground Workflows: Implemented new daily jobs and schedules within the playground environment to automatically observe source data and manage document removal processes.

Changed

⚙️ Refined Asset Automation: Adjusted automation configurations for document and data lake file removal assets, transitioning from eager conditions to more explicit job-driven scheduling.
🔗 Specialized I/O Manager: Updated the SharePoint observable asset to utilize a dedicated sharepoint_io_manager, ensuring more appropriate handling of SharePoint data interactions.

Fixed

🐛 Robust Figure Deletion: Improved the document figure deletion process to gracefully handle cases where associated figure directories do not exist, preventing unnecessary errors and log messages.

[v0.222.0] - 2025-07-15 - Agentic Processes Go Live: Full Stack Support for Complex Workflows

Added

🦾 Introduced Agentic Process Management: A major new capability enabling the definition, execution, and monitoring of complex, multi-step workflows involving humans, agents, and external programs.
🖼️ Dynamic Form Generation for Human-in-the-Loop: Processes can now define and expose forms (using Formkit elements) directly within their structure, allowing frontends to dynamically render UIs for human input, eliminating manual UI development for common interactions.
⚡️ New Process API Endpoints: A comprehensive set of new API endpoints under /processes for discovering available processes, retrieving their details, and submitting data via dynamic forms to start or continue process walkthroughs.
🔄 Dedicated Process Event Persistence: Implemented new database entities and services (PersistedProcessEventEntity, ProcessEntity) to durably store and manage all events and metadata related to agentic processes.
🚀 Real-time Process Event Streaming: Introduced new WebSocket event streams for processes, enabling real-time updates and visualization of process execution in client applications.
🏗️ Generalized Endpoint Discovery Service: A new abstract base class EndpointsDiscoveryService provides a reusable foundation for dynamically registering API endpoints based on discovered entities (like agents and processes).
🧪 Comprehensive Process Testing Infrastructure: Added SimulatedProcessApiTestRunner and enhanced ProcessTestRunner with capabilities to inject events and observe process behavior, facilitating robust testing of agentic processes.
📂 Rich Process Playground Examples: Included new, self-contained examples for various process interaction patterns (agent-only, human-only, agent-to-human, human-to-agent) to demonstrate the new capabilities.
📄 Expanded Event Specifications (EventSpecs): Generalized event schema definitions for both agents and processes, enhancing reusability and clarity in API communication.
👥 User Access to Processes in Dashboard: Users can now view their access levels to specific processes directly within their dashboard interface.

Changed

🎛️ WebSocket Event Handling Refinement: The main WebSocket endpoint (/ws) is now read-only for agent events, providing a more secure and streamlined channel for receiving real-time updates from agents. User-initiated agent events should now be sent through dedicated API endpoints.
🏷️ Improved Event Naming Consistency: Renamed WsServerEvent to ContextualizedAgentEvent and updated related API routes and client-side composables to clearly distinguish between agent and process events.
🚀 Agent Discovery Caching Enhancement: Increased the cache size for agent discovery responses, improving performance for frequently requested agent lists.
📚 API Documentation Clarity: Enhanced Pydantic annotations for various API DTOs and event models, providing more precise descriptions and examples in the OpenAPI documentation.
🌐 Frontend Localization for Dynamic Forms: Integrated Formkit's i18n capabilities to ensure dynamically rendered forms correctly reflect the user's selected locale.

Fixed

🐛 Robust Agent Retrieval: Corrected agent retrieval logic in the persistence layer to gracefully handle cases where a specific agent might not be found.
🐞 Process Event Persistence Accuracy: Resolved issues ensuring that process-related events are correctly persisted with their associated process context.
📦 NATS Topic Validation for Processes: Stricter validation for NATS process discovery topics ensures that only correctly formatted subjects are processed.
🧰 Improved Serialization of Nested Pydantic Models: Enhanced event serialization to correctly handle nested Pydantic models and ChatMessages, especially when serialize_as_any is used, preventing data loss.

Refactor

🧹 Codebase Naming Alignment: Standardized naming conventions across the codebase, particularly for "event distributors" (e.g., ExternalEventDistributor is now ExternalAgentEventDistributor), improving clarity and maintainability.
📄 Documentation Cleanup: Removed redundant or outdated comments and simplified docstrings in several core components, focusing on concise explanations.
⚙️ Pydantic Model Default Handling: Aligned Pydantic model definitions with best practices by moving default value assignments from Field annotations to direct attribute assignments.
🗂️ Test Runner Restructuring: Reorganized test runners into a more logical simulation/agent and simulation/process directory structure.
✂️ Removed Redundant Health Page: A basic placeholder health page in the frontend was removed.

Changed

🔄 Updated RAG Context Prompt Role: The chat role for the RAG context prompt has been adjusted from system to user across all supported languages (German, English, French, and Italian) to better align with prompt engineering best practices and enhance model interpretation of contextual information.

Fixed

🖼️ Corrected Image URL Handling in RAG Prompts: Resolved an issue in the English RAG context prompt where image URLs were not consistently processed, ensuring that images are now correctly referenced and displayed by explicitly converting their URL object to a string representation.

[v0.220.0] - 2025-07-11 - Enhanced SharePoint Integration Clarity and Data Lake Observability

Refactor

🧹 Standardized SharePoint Naming Conventions: Refactored variable names, function parameters, and internal asset definitions across SharePoint-related assets, ops, and IO managers for improved consistency and readability (e.g., sharepoint_ was consistently renamed to share_point_).
🔄 Updated SharePoint Asset Factories: Renamed the sharepoint_files_to_data_lake_files_factory function and its internal parameters to align with the new, standardized naming conventions, enhancing code clarity.
📄 Refined SharePoint IO Manager: Updated the SharePoint IO Manager to use the new share_point_client resource name consistently, improving maintainability.
🧼 Cleaned Up SharePoint Utilities: Standardized naming within metadata utility functions related to SharePoint files, contributing to a more coherent codebase.

Changed

⚡️ Improved Data Lake Document Parsing Logging: Added a new log statement to output the Data Lake file URI during the document parsing process, enhancing observability and assisting with debugging.

[v0.219.0] - 2025-07-11 - Internal Workflow Automation Adjustment

Removed

🗑️ Automatic Draft PR Workflow: The GitHub Actions workflow that previously created automatic draft pull requests for new branches has been removed.

[v0.218.0] - 2025-07-11 - Enhanced API Configuration and Extensibility

Added

✨ Enabled custom environment variable configuration for API services: This new capability allows users to define and inject additional environment variables, including secret references, directly into API deployments, significantly increasing configuration flexibility and integration possibilities.

[v0.217.0] - 2025-07-10 - Automated Agent API Integration and Event Inheritance

Added

✨ Automated Agent API Endpoint Generation: Introduced a new AgentDiscoveryService that dynamically discovers agents and registers corresponding API endpoints for their start events. This significantly streamlines the integration of new agents with the API, eliminating the need for manual endpoint configuration.
🦾 Event Inheritance Tracking: EventSpecs now include an event_parents field, providing a hierarchical view of event types. This new metadata enhances dynamic model generation and event handling capabilities.
📄 EventSpecs Creation Helper: Added a convenient EventSpecs.from_event_class method to simplify and standardize the generation of event specifications from event classes.
🔑 Role Provisioning for API Tokens: The generate_api_token script has been enhanced to automatically create specified roles if they do not already exist, improving the initial setup experience for new environments.
⚠️ Refined Unauthorized Access Handling: Introduced a specific HTTP exception to provide clearer and more precise error responses when a user lacks authorization to view a particular thread.

Changed

⚡️ Flexible Agent Event Dispatch: The AgentService.send_event method now universally accepts any BaseEvent as input, offering greater flexibility in initiating diverse agent interactions.
🔄 Dynamic Agent Endpoint Management: Transitioned the agent event endpoint management from static, hardcoded definitions in AgentController to a dynamic, discovery-driven approach powered by the new AgentDiscoveryService.
📚 Event Model Creation Encapsulation: Internal methods within EventModelCreationService have been refactored for improved code organization and clearer encapsulation.
🏗️ API State Access for Services: The ApiRunner now stores the AgentController and API application instance directly in FastAPI's state, enabling seamless access for new services like AgentDiscoveryService.

Refactor

🧹 Unified Event Specification Creation: Standardized the process of creating EventSpecs across AgentRunner and SimulatedAgentBotTestRunner to consistently use the new EventSpecs.from_event_class helper.
🏷️ Agent Configuration Entity Renaming: The AgentConfig embedded document within the persistence layer was renamed to AgentConfigEntity for improved naming consistency across the codebase.
✂️ Base Event Serialization Helper: Extracted common item dumping logic into a dedicated static method _item_dump within BaseEvent for improved code cleanliness.

Removed

🗑️ Static Agent Event Endpoint Configuration: Eliminated the manual and hardcoded API endpoint generation logic for agent events from AgentController and related playground examples, fully transitioning to the dynamic agent discovery service.

[v0.216.0] - 2025-07-08 - Foundational Polish: Modernizing Python Syntax and Enhancing Type Safety

Added

✨ Deprecated AgentConfig Fields: Introduced deprecation warnings for color, voice, and system_prompt fields within AgentConfig, guiding users to define these properties directly in agent subclasses for better customization.

Changed

🔄 Standardized Datetime Handling: Updated datetime operations to consistently use timezone-aware UTC for improved time-based accuracy and consistency across the platform.
🔄 Improved Subprocess Execution: Enhanced subprocess.run calls in internal scripts for more robust and secure execution, particularly within bot setup processes.
🔄 jambo Library Update: Upgraded the jambo schema conversion library to v0.4.0, bringing internal improvements for dynamic model creation and API event handling.
📄 Refined Docstrings and Comments: Various docstrings and inline comments across the codebase were updated for greater clarity and precision, especially concerning complex components like dispatchers and service methods.

Fixed

🐛 Clarified Error Messages: Enhanced error messages in several API endpoints (e.g., thread and role management) to provide more informative feedback to users.
🐛 Robust Type Parsing: Corrected and improved the internal parsing of type annotations for event and parameter extraction, ensuring more reliable and accurate workflow dispatching.

Refactor

🧹 Type Hint Modernization: Performed a comprehensive refactoring across the entire codebase to leverage native Python type hints (e.g., list, dict, set, tuple, X | None) for improved readability, maintainability, and static analysis.
🧹 Linting and Formatting Overhaul: Consolidated and enhanced code quality checks by deprecating isort and configuring ruff to handle import sorting and enforce stricter linting rules.

[v0.215.0] - 2025-07-08 - Granular Access Control and Enhanced User Management

Security

🔑 Introduced Hierarchical Role-Based Access Control (RBAC): Implemented a new, comprehensive RBAC system that supports hierarchical access rules with granular control over services, agents, and processes.
🔐 Enforced Granular Endpoint Permissions: All major API endpoints across agents, evaluations, events, files, i18n, knowledge, OpenAI, threads, and tokens now utilize the new RBAC, ensuring fine-grained access based on user roles and specific resource permissions.
🔒 Restricted Knowledge Base Access: Access to knowledge bases, including databases, documents, and nodes, is now limited to users with administrative privileges.
🛡️ Enhanced Thread Management Security: Improved security checks for thread creation, viewing, and modification, including verifying user and agent access, and preventing unauthorized changes to process-owned threads.
👁️ Refined Event Visibility: The general event retrieval endpoint has been removed, and access to thread-specific events and event timeseries is now strictly controlled by user permissions.

Added

🗄️ Role Management API: Introduced a new API (/roles) enabling administrators to create, list, retrieve, update, and delete custom roles with associated hierarchical access rules.
📋 User Management API for Administrators: New API endpoints (/users, /users/{user_id}) provide administrators with the ability to list all users and retrieve detailed information for specific users, including their assigned roles and granular access levels across services, agents, and processes.
👤 Minimal User DTO: Introduced a lightweight MinimalUserDTO for contexts where only essential user information is required, improving API efficiency.
📊 Paginated User Listing: Administrators can now retrieve a paginated list of all users via the API, enhancing user management capabilities.
👑 User Access Details in User Profile: Users can now view their specific access levels for different services, agents, and processes directly in their profile (visible to admins).
💬 Process-Associated Thread Metadata: Added process_class, process_id, and process_walkthrough_id to ThreadDTO for better contextualization of threads initiated by processes.
🤖 Agent Metadata in Model Details: agent_class and agent_id fields were added to ModelDetails to provide clearer context for AI-Hub agents exposed as OpenAI models.
⌛ User-Friendly Time Ago Display: Implemented a new composable for displaying relative time (e.g., "Last hour", "Last week") with severity indicators in the UI.
💥 Global HTTP Error Toasts: Configured the frontend client to display informative toast messages for HTTP response errors, improving user feedback.
🛠️ New UI Components for Role Management: Introduced dedicated components for creating, editing, and displaying roles, enhancing the administrative user experience.
👨‍💻 New UI Components for User Management: Added new components to list users and display their detailed access information for administrators.

Changed

🔄 Updated Role Naming Convention: The AllAgents role has been renamed to TestOnlyFullAdminAccess for clearer intent and consistency with the new access control model.
🛣️ Streamlined Frontend Routing: All previously "admin" specific routes (e.g., /admin/agents) have been unified under a /service/ prefix, aligning with the new granular access control model where permissions define access rather than a fixed "admin" route.
🌐 Refined Internationalization Endpoint Access: The /i18n/my-locale endpoint now requires a general user permission.
📂 File Access Endpoints: Access to logged-in file URLs is now managed by the new permission system.
📈 Dashboard Event Labels: Improved clarity of labels for various event types displayed in dashboard widgets.
🎨 Service Selection Layout: Adjusted the layout of the service selection component for improved visual consistency.

Refactor

🧹 Centralized Access Control Logic: All access validation logic has been consolidated into the new AccessChecker class within aihub_lib, removing redundant checks from individual controllers and DTOs.
⚙️ Refined Controller Initialization: The Controller base class now uses a more flexible additionally_required_permission parameter instead of a simple is_admin_only flag.
🚀 Optimized Service Layer Dependencies: Removed direct IdentityProvider dependencies from several service methods, pushing identity retrieval and access checks to higher layers or internal UserEntity management.
📦 Consolidated User DTOs: Simplified the user data transfer object hierarchy by replacing MyUserDTO with a more comprehensive UserDTO that includes roles and dashboard configurations, and introducing MinimalUserDTO for basic user representation.
🧪 Improved Testing Infrastructure: Enhanced test client fixtures and introduced dedicated mock fixtures for RoleEntity and UserEntity methods, making authentication and data-related tests more robust and easier to set up.
⏱️ Standardized Caching Durations: Updated staleTime configurations across multiple composables to use date-fns minutesToMilliseconds for better readability and consistency.

[v0.214.0] - 2025-07-08 - Streamlined Event Model Handling and Enhanced Developer Experience

Added

✨ New Event Model Creation Service: Introduced EventModelCreationService to provide a unified and robust mechanism for generating Pydantic models from event classes or JSON schemas, enabling dynamic and consistent API model definitions.
🧪 Comprehensive Event Model Tests: Added extensive test suites for the new EventModelCreationService, including complex nested models, union types, and list structures, to ensure the reliability and correctness of dynamically generated event models.
🛠️ Standardized IDE Inspection Profiles: Integrated new IntelliJ IDEA inspection profiles and scopes (Core.xml, ExcludeJambo.xml) to enforce consistent code quality standards and improve developer experience.
📦 Jambo Dependency: Introduced jambo as a new dependency to support advanced JSON schema to Pydantic model conversion within the new event model creation service.

Changed

🚀 CI/CD Workflow Enhancements: Updated the analyze-test-pr.yml GitHub Actions workflow to include SSH setup and cleanup steps, facilitating secure interactions with private Git repositories during testing.
🔄 API Endpoint and Test Integration: Migrated existing agent API endpoints and their associated tests to utilize the new EventModelCreationService, ensuring all model generation leverages the centralized logic.

Removed

🗑️ Deprecated Event Model Functions: Removed the standalone create_input_model and create_output_model functions, as their functionality has been absorbed and improved by the new EventModelCreationService.

[v0.213.0] - 2025-07-07 - New Retrieval Agent for Focused Data Access

Added

✨ Introduced RetrievalAgent: A new, specialized agent designed to streamline information retrieval from knowledge bases. This agent focuses solely on fetching and organizing relevant context, enabling more flexible and modular Retrieval-Augmented Generation (RAG) workflows.
📄 Dedicated Configuration for Retrieval Agent: Added RetrievalAgentConfig to provide comprehensive configuration options for customizing retrieval behavior, including specifying embedding models, vector stores, and various node retrieval strategies.
⚡️ New Event Types for Retrieval Workflows: Implemented QuestionStartEvent for clear initiation of retrieval queries and RetrievalResponseEvent to encapsulate the retrieved and ordered context, facilitating seamless integration with subsequent processing steps.
🧪 Comprehensive Test Coverage for RetrievalAgent: Included a full suite of BDD (Behavior-Driven Development) tests to ensure the robust and reliable operation of the new RetrievalAgent, validating its ability to accurately retrieve and combine relevant documents.

[v0.212.0] - 2025-07-03 - Flexible OpenAI Authentication & Resource Management

Added

✨ Flexible OpenAI Resource Authentication: Introduced the ability to authenticate with Azure OpenAI resources using either an API Key or Azure AD credentials, providing greater flexibility in deployment environments.
🔑 Centralized OpenAI API Key Configuration: Added a new base OpenaiResourceSettings class for consistent management of OpenAI API keys across different services and environments.

Changed

🔄 Generalized Resource Configuration: Extended the base ResourceConfig to include an optional api_key field, enabling API key authentication for all derived generative AI resource configurations.
🚀 Improved Azure OpenAI Client Initialization: Updated the Azure OpenAI client instantiation logic to intelligently use either an API key or Azure AD token provider based on the provided configuration, simplifying setup.
📄 Updated Development Playground Models: Renamed the o1-mini model to gpt-4o-mini in the development environment configuration for improved clarity and alignment with current model names.

Refactor

🧹 Streamlined API Key Definitions: Removed redundant api_key fields from specific model configurations (e.g., chat LLMs, embedding LLMs, image models) as they are now managed centrally by the generalized ResourceConfig.

[v0.211.0] - 2025-07-03 - Enhanced RAG Context and Workflow Streamlining

Added

✨ Introduced Parent Summary Node Retrieval: The RAG Agent can now retrieve relevant parent summary nodes from the knowledge base, providing richer context and improving the quality of generated answers.
⚙️ Configurable Parent Summary Retrieval: Added a new retrieve_summaries configuration option to the RetrieveStepConfig for RAG agents, allowing users to specify the maximum hierarchical levels for parent summary retrieval.
🚀 Improved Agent Step Metadata: Enhanced the @step() decorator in the RAG Agent to include explicit name and description fields, improving clarity and introspection for agent workflows.

Changed

⚡️ Streamlined LLM Response Workflow: The RAG Agent's LLM response step now directly signals the completion of the workflow, simplifying internal event handling and removing the need for a separate StopEvent.

[v0.210.0] - 2025-07-03 - SharePoint Ingestion and Pipeline Structure Enhancements

Added

✨ Introduced SharePoint Integration: A comprehensive new set of assets, operations, and resources for ingesting documents directly from SharePoint into the data lake. This includes robust capabilities for fetching, transforming, and managing files, ensuring your knowledge base is kept up-to-date with SharePoint content.
🚀 New Azure OpenAI Model Configurations: Added dedicated configuration classes (EmbeddingModelConfig and LanguageModelConfig) to simplify the setup and management of Azure OpenAI embedding and language models, enabling clearer model definitions.
⚙️ Flexible Pipeline Scheduling: Enhanced scheduling utilities to allow for more granular control over pipeline execution times, including a new daily_schedule_at function for precise timing.
📄 SharePoint IO Manager: Implemented a new I/O manager specifically for SharePoint, streamlining the process of loading files from SharePoint into pipeline assets.
📊 SharePoint Metadata Utilities: Introduced helper functions to generate rich metadata tables for SharePoint files, improving observability and insights into ingested data.

Refactor

🧹 Pipeline Asset Reorganization: Restructured the pipeline's asset factories into more logical subdirectories (data_lake_to_vector_store and share_point_to_data_lake) to improve modularity, readability, and maintainability.
🔄 Centralized SharePoint Utilities: Consolidated SharePoint-related operations, resources, and data types into dedicated modules for better code organization and clarity, preparing for future expansions.

[v0.209.0] - 2025-07-03 - Automated PR Readiness Checks with Claude Integration

Added

🦾 Automated Subproject PR Readiness: Introduced a new intelligent script that automatically identifies the root of a subproject based on edited files and executes make pr-ready, standardizing pre-PR checks and ensuring code quality.
⚡️ Claude Post-Edit Automation: Configured Claude to automatically trigger the new PR readiness script immediately after any file modifications (Write, Edit, MultiEdit), providing instant feedback on code quality and readiness within the development workflow.

[v0.208.0] - 2025-07-03 - Refined Context Metadata for Generative AI

Changed

📄 Streamlined RAG Context Metadata: The document_title is now explicitly included in the metadata when combining nodes for Retrieval Augmented Generation (RAG) contexts, providing more direct document identification within prompts.
🗑️ Consolidated Node Metadata Fields: To simplify and optimize the context provided to generative AI models, the namespace, type, and content_type fields are no longer emitted as part of the combined node metadata.

[v0.207.0] - 2025-07-01 - Streamlined Release Tagging

Removed

🗑️ Streamlined Release Tagging: Removed the redundant step of deleting existing Git tags in the CI/CD workflow, simplifying the release automation process.

Changed

🔄 Platform Version Alignment: All core microservices (aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, aihub_pipeline, aihub_process) and their internal aihub_lib dependencies have been updated to v0.207.0, ensuring a unified and consistent release across the entire platform.

[v0.206.0] - 2025-07-01 - Core System Modernization: Enhanced Model Definitions

Refactor

🧹 Updated Pydantic Model Definitions: Migrated numerous internal data models across all core services (aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, aihub_pipeline, aihub_process, webui_pipelines) to leverage Pydantic v2's Annotated type hint and explicit default value assignments. This enhances code clarity, type safety, and aligns the codebase with modern Python best practices.

[v0.205.0] - 2025-07-01 - Core Module Version Synchronization

Changed

🔄 Core Module Version Alignment: All internal microservices (aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, aihub_pipeline, aihub_process) and their aihub_lib dependencies have been updated to v0.205.0, ensuring consistent versioning across the platform.
⚙️ Build Tag Update: The default build tag in the Makefile has been updated to v0.205.0 to reflect the latest release version.

[v0.204.0] - 2025-07-01 - Release Workflow Enhancements and Changelog Automation Robustness

Added

✨ Robust Changelog Generation: Implemented error handling within the changelog generation script, allowing the release process to continue gracefully even if the Large Language Model fails to produce an entry.
📝 Detailed Changelog Logging: Enhanced logging in the CI/CD pipeline to provide clearer feedback on whether the changelog.md file was successfully updated or skipped during the automated generation process.

Changed

⚡️ CI/CD Workflow Resilience: The automated tag and release workflow now allows the changelog generation step to continue-on-error, prioritizing the overall release completion over a potential failure in AI-powered changelog creation.
🔄 Conditional Changelog Commit: The final commit of the changelog.md file is now dependent on the successful execution of the changelog generation step, preventing the amendment of the release commit with an incomplete or erroneous changelog.
⚙️ Gemini API Key Command: Updated the command-line syntax for setting the Gemini API key in the CI/CD pipeline to use the --value flag, aligning with the llm tool's latest requirements.

Refactor

🧹 Python Poetry Cache Removal: Streamlined the GitHub Actions Python setup by removing the explicit Poetry cache configuration, potentially improving caching efficiency or addressing previous issues.

[v0.203.0] - 2025-06-30 - Automated Changelog Generation & Enhanced Release Workflow

Added

✨ Automated Changelog Generation: Introduced a new system to automatically generate changelog entries for releases, leveraging an LLM to summarize git diffs. This includes a dedicated script (generate-changelog.sh), an LLM prompt file (changelog-prompt.md), and a new make changelog target.
📄 Centralized Changelog File: A new changelog.md file is now part of the repository, serving as the official documentation for all notable changes in a standardized format, automatically updated during the release process.

Changed

🚀 Enhanced Release Automation Workflow: The GitHub Actions workflow (add-tag.yml) has been significantly upgraded to seamlessly integrate the new automatic changelog generation, ensuring comprehensive release notes are created and committed as a core part of the tagging and release process.
⚙️ Streamlined CI/CD Setup: Improved the efficiency and robustness of Python and Poetry setup within the CI/CD pipeline, including a more reliable method for installing Poetry via pipx and consolidating global dependency installation steps.
🔄 Refined Commit and Tagging Logic: The release workflow now intelligently handles version updates and changelog changes, amending the previous commit with generated changelog entries and ensuring more robust force pushes for tags.

Refactor

🧹 Optimized CI/CD Workflow Steps: Internal steps within the add-tag.yml workflow have been reorganized and renamed for better clarity and maintainability, streamlining the automated release pipeline.

[v0.202.0] - 2025-06-30 - Unified Identity & Auth: Stronger, Smarter, Safer

Refactor

🧹 Unified User Identity Management: Replaced the disparate AuthenticatedUser model and various API-specific user information providers with a new, centralized UserIdentity model and a consistent IdentityProvider framework within aihub_lib. This significantly improves modularity, consistency, and extensibility for handling user identities across all services (aihub_api, aihub_bot).
🧹 Streamlined Authentication Handlers: Refactored core authentication handlers (AuthHandler, OAuth2AuthHandler, TokenAuthHandler, OpenWebuiAuthHandler, TokenAndOauth2Handler) to seamlessly integrate with the new IdentityProvider framework, simplifying their internal logic and promoting dependency injection for clearer authentication flows.
🧹 Renamed Development-Only Authentication: Renamed NoAuthHandler and its associated configurations to DangerousDevelopmentOnlyAuthHandler and DangerousDevelopmentOnlyAuthConfig respectively. This clearer naming strictly communicates that these components are intended for development and testing environments only, preventing their accidental use in production.

Changed

🔄 Updated User Representation in API & UI: The preferredUsername field in various API responses and UI components (such as chat messages and event displays) has been consistently renamed to email to more accurately reflect the user's email address.
⚙️ API Controllers Initialization Refinement: All API controllers now explicitly require an AuthHandler during initialization, enforcing clearer dependency management and simplifying security configuration across the application.
📄 Refined Image Description Prompting: The internal prompt used for generating figure descriptions has been updated with more precise instructions on how to leverage surrounding text for context without altering the factual, visual description of the image.

Fixed

🐛 Improved Figure Description Generation Robustness: Enhanced the image description generation process in the data pipeline with robust error handling for BadRequestError (often triggered by content safety filters). The system now automatically retries generating descriptions without surrounding text and falls back to an empty string if necessary, preventing pipeline failures.

Added

✨ Introduced by_email User Lookup: Added a new by_email class method to UserEntity, enabling direct and efficient retrieval of user entities by their email address from the database.
🚀 Enhanced OAuth2 Token Verification Caching: Implemented internal caching mechanisms for JWKS (JSON Web Key Sets) and RSA keys within OAuth2AuthHandler, significantly improving the performance and reducing external API calls during OAuth2 token validation.

Removed

🗑️ Deprecated AuthenticatedUser Model: The AuthenticatedUser data model has been entirely removed, succeeded by the more versatile and centrally managed UserIdentity model.
🗑️ Removed Redundant Identity Provider Implementations: Obsolete, API-specific identity provider implementations (ApiTokenUserInformationProvider, AzureUserInformationProvider, DevUserInformationProvider) have been removed, replaced by the new, unified IdentityProvider framework in aihub_lib.
🗑️ Unauthenticated Health Endpoint: The health endpoint no longer requires user authentication, simplifying its use for external readiness and liveness probes.

[v0.201.0] - 2025-06-30 - Unified User Identity & Authentication Overhaul

Refactor

🔄 Unified User Identity Management: Implemented a comprehensive overhaul of user identity handling, centralizing the UserIdentity model in aihub_lib and replacing the deprecated AuthenticatedUser across all services (aihub_api, aihub_agent, aihub_bot). This standardizes how user information is represented and accessed.
🧹 Decoupled Authentication from Identity Resolution: Introduced a new IdentityProvider interface and various concrete implementations (e.g., AzureIdentityProvider, TokenIdentityProvider). Authentication handlers (OAuth2AuthHandler, TokenAuthHandler, OpenWebuiAuthHandler) now leverage these providers, significantly improving modularity and allowing for easier integration of new identity sources.
🔐 Enhanced Security Development Experience: Renamed NoAuthHandler to DangerousDevelopmentOnlyAuthHandler and introduced a dedicated DangerousDevelopmentOnlyIdentityProvider to clearly delineate and manage authentication in non-production environments, making it more explicit when security is bypassed for development.

Changed

⚡️ Standardized Controller Initialization: The auth parameter in all Controller constructors is now a required keyword-only argument, promoting clearer and more consistent API initialization.
🚀 Unauthenticated Health Checks: The health check endpoint (/health) no longer requires authentication, making it easier to monitor service availability.
📄 API Token Generation Improvements: Updated the generate_api_token.py script with clearer CLI parameter names (--token-path, --token-url) to better reflect their purpose.
🛣️ Bot API Path Update: Changed the base API path for bot-in-the-loop interactions from /api/v1 to /bearer_token/v1 in aihub_bot to align with specific authentication routing.
🌐 Frontend User Profile Update: Renamed the preferredUsername property to email in frontend user components and SDK schemas to align with the new UserIdentity model, reflecting a more direct and generic user email representation.

Removed

🗑️ Deprecated User Information Providers: Eliminated several specific user information provider classes from aihub_api (BaseUserInformationProvider, MultiStrategyUserInformationProvider, ApiTokenUserInformationProvider, AzureUserInformationProvider, DevUserInformationProvider) as their functionality is now superseded by the new IdentityProvider architecture in aihub_lib.
🗑️ Legacy AuthenticatedUser Model: The outdated AuthenticatedUser model has been fully removed from aihub_lib, completing the transition to UserIdentity.

Added

🦾 Introduced Agentic Processes (aihub_process): A new top-level module is added to enable sophisticated orchestration of work across diverse entities, including AI agents, human collaborators, programs, and even other processes. This foundational change allows for building complex, multi-entity workflows.
⚡️ Generic Dispatcher (BaseDispatcher): A new abstract base class BaseDispatcher is introduced in aihub_lib, centralizing common logic for handling events, managing execution contexts, and executing workflow steps across all dispatchable entities (agents and processes).
⚙️ Generalized Event and Step Stores: Core event storage (JetStreamEventStore) and step execution tracking (StepStore) are moved and refactored into aihub_lib, providing a unified, distributed persistence layer for all workflow types.
🔗 New Event Types for Process Communication: A comprehensive set of WorkEvent and WorkRequestEvent hierarchies are introduced (AgentWorkEvent, HumanWorkEvent, ProgramWorkEvent, ProcessWorkEvent and their WorkRequest counterparts) to standardize how different entities interact within a process.
📚 Process-Specific Topics and Subscribers: Dedicated topic managers (ProcessTopicManager, ProcessInstanceTopicManager, ProcessWalkthroughTopicManager) and NATS subscribers (ProcessJSSubscriber, ProcessNCSubscriber) are added to define and manage communication channels for process-specific events.
📝 Process Configuration (ProcessConfig): A new configuration class is introduced to define metadata and settings for process instances, enabling flexible deployment and management.
🌐 Process Locale Support: New localization files are added to support process-related terms and messages.

Changed

🔄 Unified Workflow Base (DispatchableWorkflow): The Agent class now inherits from DispatchableWorkflow, establishing a common architectural pattern for all event-driven entities and centralizing step discovery and event handling logic.
🧹 Refactored Agent Dispatcher: The AgentDispatcher is renamed and refactored to extend the new BaseDispatcher, streamlining its implementation by leveraging shared core dispatcher functionalities.
🧩 Modularized NATS Communication Components: TopicManager, JSPublisher, and NCPublisher are refined into more generic base classes, while agent-specific logic is migrated to new, dedicated AgentTopicManager and AgentNCSubscriber/AgentJSSubscriber classes for better modularity.
🛠️ Updated CI/CD and Build Configuration: GitHub Actions workflows and the main Makefile are updated to incorporate linting, testing, and deployment steps for the new aihub_process module.
📦 API and Bot Module Alignments: The aihub_api and aihub_bot modules are updated to utilize the new ExternalAgentEventDistributor and other agent-specific communication components, ensuring compatibility with the refined messaging architecture.
🎨 Enhanced LocaleString Flexibility: display_name and display_description fields in display events now support None values in LocaleString, providing greater flexibility in event schema definition.
🚧 Agent Configuration Enhancements: Improved validation for agent_id with regex pattern constraints. color, voice, and system_prompt fields in AgentConfig are deprecated, encouraging more granular, domain-specific configuration for agents.
📄 Documentation Updates: The agent documentation is updated to reflect the AgentDispatcher renaming and the new architectural patterns.

Fixed

🐛 API Typecheck Issue: Corrected a typo in the aihub_api/Makefile to ensure proper type checking for the aihub_api module.
🔒 Database Uniqueness Constraints: Added unique=True indexes to conversation_id in ConversationEntity and path in PathEntity within aihub_bot, and email in UserEntity to enforce data integrity and improve query performance.

Refactor

🧹 Consolidated Context Management: RunContext and ThreadContext are moved to aihub_agent/context, making their scope agent-specific and aligning with the new modular structure.
🏷️ Explicit External Event Naming: ExternalEvent and ExternalEventDistributor are renamed to ExternalAgentEvent and ExternalAgentEventDistributor respectively, clarifying their specific role in handling external communication for agents.
🧹 Streamlined Type Extraction: The extract_event_classes utility is enhanced to properly unwrap Annotated types, ensuring more accurate type introspection for workflow steps.
🗑️ Cleaned Up Topic Hierarchy: The base Topic class is made abstract, and TopicManager is stripped down to its core, generic functionality, with all agent-specific topic logic moving to the new AgentTopicManager and its subclasses.

[v0.199.0] - 2025-06-27 - Enhanced Document Processing and New Docling Integration

Added

✨ Introduced Docling Document Loader: Added a new DoclingLoader to integrate with the Docling API, expanding support for parsing various document types including PDFs, office documents, and images.
🦾 LLM-Powered Figure Description Generation: Implemented a new pipeline operation (generate_figure_descriptions) that uses Large Language Models to automatically create detailed, accessible alt-text descriptions for figures within documents.
⚙️ Configurable Document Parsing Engines: Introduced LoaderType in DocumentParserResource, allowing explicit selection between Docling, Azure Document Intelligence, or using both for document processing, providing greater flexibility.
📚 Docling API Integration: Added core components (DoclingAccess and DoclingConfig) for seamless interaction and configuration with the Docling document conversion API.
📄 Expanded Document Intelligence Supported Formats: Configured the DocumentIntelligenceLoader with an explicit list of supported file extensions, enhancing its capability to parse various document formats.

Changed

🔄 Streamlined Document Intelligence Processing: The DocumentIntelligenceLoader now directly handles table reformatting and figure extraction/saving, simplifying the document processing flow by embedding this logic within the loader itself.
💬 Updated Figure Description Prompts: Revised the prompt structure for figure description generation to better leverage multi-modal LLMs and improve the quality of generated alt-text.
📝 Simplified Figure Metadata: The FigureMetadata model has been streamlined by removing the figure_url field, focusing on essential file path information.
🚧 Playground Default Loader: Updated the playground environment to default to the DoclingLoader, showcasing the new integration.

Refactor

🧹 Optimized Document Processing Pipeline: Significantly refactored the document processing asset graph by removing redundant intermediate operations and consolidating logic directly within document loaders for a more efficient and cleaner pipeline.
📦 Reorganized Path Utilities: Moved and renamed the path_utils module from aihub_pipeline to aihub_lib and updated function names for better modularity and broader utility across the aihub_lib.

Removed

🗑️ Deprecated Document Processing Operations: Eliminated several no-longer-needed pipeline operations, including doc_with_figures_to_ref_doc, inject_figures, reformat_tables, and save_figures_to_data_lake, as their functionalities have been integrated directly into document loaders.

[v0.198.0] - 2025-06-24 - Pipeline Embedding Enhancements

Changed

🦾 Improved node embedding content within pipelines by transitioning from get_text() to get_content(metadata_mode=MetadataMode.EMBED) for extracting text. This ensures that relevant metadata is also considered, leading to potentially richer and more accurate vector embeddings.

[v0.197.0] - 2025-06-24 - Enhanced Infrastructure with Azure Data Lake Storage Support

Added

✨ Introduced Azure Data Lake Storage Management: Added new Infrastructure as Code (IaC) components to provision and manage Azure Data Lake Storage accounts, expanding the platform's data storage capabilities.
🔑 New Azure Data Lake Roles: Incorporated specific Azure role definitions (Storage Blob Data Reader and Storage Blob Delegator) to facilitate secure and controlled access to data within Data Lake Storage.
🦾 Automated Data Lake Role Assignment: Implemented a new method within UserAssignedIdentity to automatically assign the necessary permissions for Data Lake Storage, specifically enabling the generation of Shared Access Signatures (SAS) tokens for temporary data access.
📄 Standardized Data Lake Naming: Defined a consistent naming convention for Azure Data Lake Storage resources, improving resource organization and clarity within the infrastructure.

[v0.196.0] - 2025-06-20 - Automation Enhancements and Version Synchronization

Added

✨ Enhanced Automated Pull Request Creation: Improved the CI/CD workflow for generating automatic draft pull requests by incorporating full Git history fetching and enabling automatic population of PR descriptions from commit messages, streamlining development processes.

Changed

🔄 Project Version Synchronization: Aligned all microservices and core libraries, including aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, and aihub_pipeline, to version v0.196.0 for consistent dependency management and release coordination.

[v0.195.0] - 2025-06-20 - Enhanced Pipeline Efficiency

Changed

⚡️ Optimized Dynamic Partition Management: The replace_partition_keys utility in the pipeline core has been significantly improved to efficiently add and delete only the necessary dynamic partitions, thereby reducing unnecessary operations and enhancing overall performance for data pipeline execution.

[v0.194.0] - 2025-06-19 - Core System Alignment and Pipeline Operation Streamlining

Changed

🔄 Core Service Version Alignment: Updated internal dependency versions across all aihub_ services (agent, api, bot, iac, lib, pipeline) and the Makefile to v0.194.0 for consistent build and deployment.

Refactor

🧹 Simplified Data Lake File Fetching: Refactored fetch_all_files_in_data_lake operations within the pipeline module to remove direct OpExecutionContext dependency and internal logging, promoting cleaner, more focused data retrieval logic.

[v0.193.0] - 2025-06-19 - Enhanced Draft PR Automation with Structured Branching

Added

✨ New Branch Naming Conventions: Implemented strict base/scope naming validation for new branches, ensuring consistency and clarity. This includes a predefined list of allowed scope prefixes.
🔒 Improved GitHub App Token Usage: Switched to using a dedicated GitHub App Token for creating draft pull requests, enhancing security and permissions management for automated workflows.
🚀 Duplicate PR Prevention: Added a check to prevent the creation of redundant draft pull requests if one already exists for the branch, streamlining workflow execution.

Changed

⚙️ Dynamic Draft PR Titles: Automated draft pull requests now generate more informative titles, incorporating the parsed base and scope from the branch name, along with the author's username, for immediate context.
➡️ Default Base Branch for Draft PRs: Standardized the base branch for automatically created draft pull requests to main, simplifying the initial setup for new feature or bugfix branches.
🧹 Streamlined Workflow Authentication: The workflow now utilizes a more secure and efficient method for authenticating with the GitHub API.

[v0.192.0] - 2025-06-18 - Streamlined Branch Naming and Project Alignment

Changed

🔄 Relaxed Branch Naming Conventions: Updated the semantic PR GitHub workflow to allow more flexible and generic branch names (e.g., feature/my-feature, bugfix/issue-123) by replacing specific initiative branch patterns with a broader regular expression. This simplifies development workflows and enhances consistency.
⬆️ Project Version Alignment: Aligned all internal project and library dependencies across microservices (aihub_agent, aihub_api, aihub_bot, aihub_iac, aihub_lib, aihub_pipeline) to the new v0.192.0 release tag, ensuring consistent versioning throughout the codebase.

[v0.191.0] - 2025-06-18 - LLM Event Reporting Enhancements

Added

✨ Improved LLM Event Reporting: Introduced a new from_chat_response method to LLMEvent for easier creation of LLM events directly from LlamaIndex chat responses, automatically capturing token usage for enhanced observability.

[v0.190.0] - 2025-06-16 - Core Updates and Path Robustness

Fixed

🐛 Enhanced Data Lake Path Generation: Improved the utility for creating data lake figure folder names by transitioning to a more robust file name parsing method (os.path.splitext), ensuring accurate and reliable folder structures regardless of complex file naming conventions.

[v0.189.0] - 2025-06-16 - Multimodal RAG and Enhanced File Access

Added

✨ Multimodal RAG Capabilities: The RAG Agent now supports incorporating images alongside text in its context, enabling it to answer questions that require visual information. This includes the dynamic generation of secure, time-limited URLs for embedded figures.
🔗 Extended Anonymous File Link Lifetime: Anonymous file sharing links can now be configured to remain valid for up to 30 days, providing increased flexibility for long-term access to shared content.

Changed

🦾 Improved RAG Agent System Prompt: The RAG Agent's core instructions have been refined to enhance its ability to analyze and provide detailed answers using both document and image context, and to explicitly state when it cannot find an answer.
🔄 Standardized Anonymous File Access: The generation and validation of secure, temporary file URLs have been centralized and streamlined within the AnonymousFileAccessService, improving consistency and security for file access.
📄 Enhanced Document Parsing for Multimodal Content: The Markdown parser now intelligently identifies and categorizes content within documents as text, figures, or tables, enriching the metadata for RAG processing.
🧩 Updated RAG Node Data Model: The IngestedNode data model now includes explicit content_type fields (text, figure, table) to support the new multimodal RAG capabilities, ensuring detailed content classification.
📝 Modernized RAG Context Prompt Templating: RAG context prompts now leverage a more advanced templating system (RichPromptTemplate) to natively support the inclusion of both text and image blocks, improving prompt construction and multimodal reasoning.
⚙️ Refined Figure and Table Injection in Pipelines: Document processing pipelines now inject figures and tables into markdown using specific XML-like tags (<figure> and <table>), enabling the downstream RAG system to correctly identify and process these content types.
🧹 Centralized ChatMessage Serialization: Serialization logic for ChatMessage objects, particularly for handling URLs in image and audio blocks, has been consolidated into the BaseEvent model for improved consistency across all events.

Refactor

📦 Reorganized File Access Logic: The internal logic for generating and validating anonymous file signatures and SAS tokens has been refactored and moved to dedicated services for better module separation and maintainability.

[v0.188.0] - 2025-06-13 - Advanced Document Handling: Secure Access & Improved Markdown Rendering

Added

✨ Secure File Access API: Introduced new API endpoints (/file) for secure access to files stored in Azure Blob Storage, supporting both logged-in users and temporary anonymous sharing via signed URLs.
📄 Azure Blob Storage Integration: Added core services and configuration for seamless integration with Azure Blob Storage, enabling centralized and secure file management across the platform.
🚀 Anonymous File Access Service: Implemented a new service to generate secure, time-limited URLs for anonymous file sharing, enhancing the platform's ability to share documents securely.
🖼️ Markdown Rich Content Components: Introduced dedicated Vue components (MarkdownFigure, MarkdownTable) to significantly improve the rendering and styling of figures and tables within markdown content.
🔗 Dynamic Image Resolution for Markdown: Added a ResolveImageComponent to dynamically fetch and display images from secure internal storage within markdown content using signed URLs.
⬇️ Document Download Capability: Enabled users to download original documents directly from the knowledge base list in the web interface, improving usability and accessibility.
📝 File Attachment Support: Extended API schemas to allow attaching files to Metadata and UserMessageEvent for richer interaction capabilities with agents.

Changed

🔄 Event Schema Refinement: Refactored event schemas into Readable and Writable types to provide clearer distinctions between data models for API input and output, enhancing type safety and API clarity.
📊 Knowledge Base Document Display: Adjusted the web interface display for knowledge base documents to prominently feature the number of pages and improved column ordering for better readability.
📄 Figure Reference Handling: Updated the document processing pipeline to handle figure references using file paths instead of direct URLs, aligning with the new on-demand URL generation for secure access.
🎨 Markdown Styling Enhancements: Refined markdown rendering styles, including adjusted heading margins and link appearances, for a more consistent and polished visual experience.
📚 Ingested Document Metadata: Enhanced the IngestedDocument model to accurately capture and display the number_of_pages from source document metadata, providing more comprehensive document details.
📦 New Dependency: Added lxml to the pipeline dependencies, which is likely used for improved parsing and processing of document structures, supporting richer content extraction.

Refactor

🧹 Azure Storage Access Standardization: Centralized Azure Blob Storage access through a new singleton BlobStorageAccess class, ensuring consistent and maintainable interactions with storage services.

Fixed

🐛 API Base URL Configuration: Ensured correct API base URL configuration for the web client, standardizing API access and preventing potential routing issues.

Fixed

🐛 Document Parsing Robustness: Enhanced the Markdown structural node parser to gracefully handle empty content, preventing potential processing errors during document ingestion.
🛡️ Table Reformatting Stability: Improved the document table reformatting process within the pipeline to prevent errors when encountering documents with missing or empty text content.

Changed

⚙️ Pipeline Function Clarity: Updated function calls within the documents_factory in the pipeline to consistently use keyword arguments, improving code readability and maintainability.

[v0.186.0] - 2025-06-12 - Enhanced Data Flexibility and Azure AI Integration

Changed

🔄 Improved Document Metadata Handling: The IngestedBase and IngestedNode document types now dynamically capture and retain all additional, non-standardized metadata fields present in source documents, significantly enriching the information available for RAG pipelines.
🔑 Enhanced Azure OpenAI Embedding Authentication: Explicitly enabled Azure Active Directory (AAD) authentication for Azure OpenAI embedding models, providing more secure and robust integration with Azure environments.

[v0.185.0] - 2025-06-11 - Internal Updates and CI/CD Streamlining

Refactor

🧹 Optimized Build Process: Removed the explicit code checkout step within the build_image GitHub Action, streamlining the image building process and potentially reducing redundant operations.

[v0.184.0] - 2025-06-11 - Web UI Streamlining and Configuration Updates

Changed

🔄 Web UI Default Development Port: Updated the default local development port for the Web UI from 5173 to 8080 in docker-compose-webui.yml, affecting local setup and direct access URLs.
⚙️ OIDC Configuration Management: Removed hardcoded OIDC client and tenant IDs from nuxt.config.ts, indicating a shift towards more dynamic or externalized authentication configuration.

Refactor

🧹 useRouteQuery Hook Usage: Refactored the usage of useRouteQuery in Web UI composables and pages to explicitly pass route and router instances, aligning with recommended practices and improving compatibility.

Removed

🗑️ Storybook Integration: Completely removed Storybook development tooling and all related configurations from the web UI project, streamlining the frontend development environment.
🗑️ Deprecated shad Makefile Command: Removed the shad command from the aihub_web Makefile, simplifying build scripts.

[v0.183.0] - 2025-06-11 - Smarter Document Summarization and Enhanced Data Traceability

Added

📄 Introduced document_store_name Metadata: Nodes now include document_store_name metadata, enhancing traceability and organization by indicating their origin.
🔗 Integrated document_store_name into Chunking Pipeline: The document chunking process now automatically populates the document_store_name metadata field for all generated nodes.

Changed

🧠 Revamped Recursive Summary Parser: A major overhaul of the recursive summary parser to generate more logical and accurate document summaries. This includes improved content grouping, robust handling of sequential sections, and precise relationship management between original and summary nodes.
🌳 Enhanced Summary Node Relationships: Summary nodes now maintain a more coherent hierarchical context within the document structure, including new PREVIOUS and NEXT pointers, and refined PARENT/CHILD relationships.
📝 Improved Summary Prompting: The underlying prompt for summarization has been updated across multiple languages (DE, EN, FR, IT) to generate more concise, direct, and "TL;DR" style summaries.
⚡️ Updated Playground Environment Configuration: Internal variable names have been updated for clarity, and default settings now utilize "papers" for data lake containers and namespaces, streamlining demonstration and testing.
🚀 Switched Playground LLM to gpt-4o-mini: The default language model for the playground environment has been updated to gpt-4o-mini, potentially offering more cost-effective or faster experimentation.

Fixed

🐛 Improved Empty Text Handling in Summarizer: The LLM summarizer now gracefully handles empty or whitespace-only input, returning an empty string to prevent potential errors or irrelevant output.

Refactor

🧹 Cleaned RecursiveSummaryParser Internals: Streamlined internal logic by integrating previously separate helper methods directly, improving code readability and maintainability.

[v0.182.0] - 2025-06-11 - Enhanced Bot Responsiveness

Fixed

🐛 Resolved Bot Typing Behavior: Corrected the timing of the bot's "typing" indicator to ensure it only appears when the bot is actively processing a response, preventing misleading indicators for interactions the bot is configured not to respond to (e.g., expired conversations).

[v0.181.0] - 2025-06-10 - Core Enhancements and Stability Improvements

Changed

🚀 Enhanced Agent Event Distribution: The AgentCompletionHandler now supports a custom display_id for agent events, providing greater flexibility in event tracking and display.
💬 Flexible OpenAI Completion Parameters: OpenAI completion methods (get_completion, get_stream_completion) have been extended to accept additional keyword arguments, allowing for more diverse model interactions.
🌍 Localized Error Handling: Integrated LocaleHandler into the OpenaiCompletionHandler's exception handling, paving the way for more user-friendly and localized error messages.

Fixed

🐛 Improved Content Type Robustness: Added checks for file_info.content_type in the ContentExtractor to prevent potential errors when processing file attachments with undefined or missing content types.

Refactor

🧹 Streamlined Completion Handler Calls: Removed redundant service parameters from get_stream_completion and get_completion calls within the BaseChatBot, simplifying internal API usage.
🔄 Centralized Completion Handler Import: Adjusted import paths for the CompletionHandler to reflect its dedicated module, improving code organization and maintainability.
⚙️ Standardized Channel ID Usage: Replaced a hardcoded string with the Channels.webchat constant in StreamAgentChatBot for improved consistency and maintainability.

[v0.180.0] - 2025-06-10 - Enhanced Bot Robustness and Core Version Alignment

Fixed

🐛 Improved Chat Bot Locale Handling: Addressed an issue in the BaseChatBot to gracefully handle scenarios where user locale information might be missing, ensuring robust locale resolution by falling back to a default.

[v0.179.0] - 2025-06-10 - Next-Gen Chat: File Uploads & Seamless WebUI Integration

Added

✨ Introduced comprehensive file upload support for chat completion requests, enabling richer contextual interactions with AI agents.
📝 Defined a new **UserUploadedFile** data model to standardize the representation of uploaded files across the platform.
🌐 Launched a new Open WebUI pipeline that seamlessly integrates file sending to AI-Hub assistants and visualizes retrieved sources within the UI.

Changed

🔄 Extended core chat services (ChatService and OpenaiService) to process and manage user-uploaded files within conversations.
⚡️ Enhanced internal event handling to propagate UserUploadedFile data, ensuring consistent context across agents and services.

Fixed

🛠️ Addressed a compatibility issue in the API token generation script (generate_api_token.py) for improved reliability with Azure CLI.

[v0.178.0] - 2025-06-05 - Infrastructure Control and WebUI Scalability

Added

✨ Introduced webui_secret_key: A new secret key can now be configured for OpenWebUI, enhancing security for the user interface.
🚀 Configurable WebUI Scaling: Enabled configuration of minimum and maximum replicas for the WebUI container, allowing for more flexible resource allocation and scalability.

Changed

⚙️ Azure Container Apps Subnet Delegation: Implemented subnet delegations for Dagster and WebUI modules, improving integration with Azure Container Apps managed environments.
💾 Dagster PostgreSQL Data Retention: Configured the Dagster PostgreSQL database to retain data upon deletion, preventing accidental data loss during infrastructure changes.
⚡️ WebUI Performance Optimizations: Enhanced WebUI performance by dynamically setting UVicorn workers and thread pool size based on allocated CPU, and integrated the new WEBUI_SECRET_KEY and Redis URL.
📊 Managed Environment Workload Profiles: Defined explicit workload profiles for Azure Container Apps, optimizing resource allocation and cost efficiency.

Refactor

🧹 Codebase Cleanup: Removed several unused imports across Agent, API, Bot, and WebUI configurations for cleaner and more efficient code.

[v0.177.0] - 2025-06-04 - Enhanced Document Processing with AI-Powered Figure & Table Management

Added

✨ Automated Figure Description Generation: Introduced AI-powered generation of detailed alt text for figures in documents, leveraging vision models to describe visual content for improved accessibility.
🖼️ Figure and Table Extraction in Document Processing: Implemented robust extraction and handling of figures and tables from documents, allowing them to be processed as distinct, enriched content elements.
📄 Page Number Metadata: Added functionality to extract and store page numbers as metadata for document nodes, significantly improving context and precise referencing within documents.
🗑️ Figure Cleanup for Deleted Documents: Introduced automatic deletion of associated figure data from the data lake when reference documents are removed, ensuring data consistency and storage efficiency.
⚡️ Prompt for Figure Description Generation: Added new multi-language prompts specifically designed for generating high-quality, accessible descriptions for figures.

Changed

🔄 Enhanced Document Processing Pipeline: Significantly restructured and enhanced the document ingestion pipeline to support comprehensive figure and table extraction, reformatting, and injection of AI-generated descriptions.
📈 Improved Table Reformatting: Updated the pipeline to convert HTML tables within documents to Markdown format, improving readability and data consistency for processed content.
🗣️ Summarizer Prompt Language Consistency: Modified summarization prompts to explicitly ensure the output summary is generated in the same language as the input text.
⚙️ Data Lake Resource Configuration: Updated data lake resources to include dedicated directories for storing extracted figures, streamlining content organization.

Refactor

🧹 Centralized Azure Suffix Constants: Moved various Azure resource suffix constants into a single, dedicated constant file for improved maintainability and consistency across Infrastructure as Code (IAC) modules.
🗂️ Refactored Document Parsing Logic: Reworked the core document parsing operations to support the new figure and table extraction capabilities, separating concerns more clearly for better modularity.

Fixed

🐛 Typo in Document Store Removal Logic: Corrected a minor typo in the logic for fetching reference documents to be removed from the document store.

[v0.176.0] - 2025-06-04 - Introducing Advanced Document Intelligence with Visual Content Processing

Added

🖼️ Multimodal Document Processing: Introduced the capability to extract figures from documents, generate detailed alt-text descriptions using an AI vision model (GPT-4o), and re-inject these descriptions and image URLs back into the document content.
⚡️ Table Reformatting: Added functionality to automatically convert HTML tables within documents into a structured Markdown format for improved readability and processing.
📄 Page Number Metadata: Document nodes now include a page metadata field, enabling precise contextual retrieval based on original page numbers.
✨ Dedicated Figure Storage: Implemented a dedicated directory in the data lake for storing extracted raw figure data, ensuring efficient management of visual assets.
🦾 New Prompt Templates for Figure Descriptions: Added specialized prompt templates to guide the AI vision model in generating high-quality, accessible alt-text for figures.

Changed

🔄 Revamped Document Processing Pipeline: The entire document ingestion pipeline has been significantly refactored to support multimodal capabilities, including new stages for figure extraction, description generation, and table reformatting.
🧠 Upgraded Language Model: The default language model for the playground environment has been upgraded to gpt-4o, enhancing its capabilities, especially for vision-related tasks.
🧹 Refined Data Lake File Handling: Updated the logic for fetching files from the data lake to exclude figure directories, streamlining document processing.
🌍 Improved Summarizer Localization: The summarizer prompt now explicitly instructs the model to generate summaries in the same language as the input text, improving language consistency.

Refactor

⚙️ Modularized Document Parsing: The core document parsing operation was refactored to separate the initial parsing from the final RefDocDocument creation, improving pipeline clarity and flexibility for multimodal steps.
🗑️ Automated Figure Cleanup: Implemented automatic deletion of associated figure data from the data lake when a document is removed, ensuring data hygiene.

[v0.175.0] - 2025-06-02 - Refined OpenAI Integrations and Infrastructure Control

Refactor

🦾 Improved OpenAI API Integration: Centralized and streamlined how request parameters are handled for OpenAI SDK calls (Chat Completions, Image Generation, Text-to-Speech). This enhancement intelligently filters and prepares arguments from Pydantic models, ensuring robust and error-free communication with the OpenAI API.
⚡️ Enhanced Internal API Consistency: API controllers now directly pass full Pydantic request models to service layers, improving type safety and internal consistency across the codebase.

Changed

🔄 Updated Core Docker Images: Switched docker-compose configurations to utilize custom-built ghcr.io images for PostgreSQL with PgVector and Open WebUI, providing greater control and consistency over deployment environments.

Refactor

🧹 Centralized Document Store Suffix: Extracted the DEFAULT_DOCSTORE_SUFFIX constant from the StoresConfig module into a new, dedicated constants file, enhancing code organization and maintainability within the infrastructure components.

[v0.173.0] - 2025-05-30 - Infrastructure Enhancements: Cosmos DB Document Store Support

Added

✨ Cosmos DB Document Store Integration: Introduced full support for provisioning and managing Cosmos DB as a document store within the Azure Infrastructure as Code (IaC) layer.
🚀 Standardized Cosmos DB Naming: Implemented a new utility to ensure consistent and predictable naming conventions for Cosmos DB document store instances.
🔑 Automated Identity Permissions for Cosmos DB: Enhanced user-assigned identity management to automatically assign the necessary roles for secure interaction with Cosmos DB document stores.

Added

✨ Comprehensive LLM Evaluation Suite: Introduced a new, full-fledged module for evaluating Large Language Models (LLMs) and agents. This includes dedicated API endpoints, backend services, LLM-based judging (via Arize Phoenix), and a complete user interface.
🦾 LLM-Powered Evaluation Judges: Integrated an LLM-powered judging mechanism for automated evaluation of agent responses on dimensions such as correctness, completeness, and conciseness, configurable via internationalization (i18n) prompts.
📄 Multilingual Evaluation Prompts: Added new i18n translations for LLM judge prompts, enabling multilingual evaluation capabilities for agent responses and ensuring broader applicability.
🚀 Arize Phoenix Integration: Integrated Arize Phoenix as the core framework for managing and visualizing LLM evaluation datasets and experiment results, providing powerful insights into agent performance.
📚 New UI Components for Evaluation: Introduced a suite of new user interface components to support the evaluation module, including intuitive cards for datasets and experiments, and detailed results views for comprehensive analysis.
⚙️ New UI Composables for Evaluation: Added dedicated Vue composables to streamline data fetching and mutation for the new evaluation features, simplifying frontend development and maintenance.
📦 Core Dependency: arize-phoenix: Incorporated arize-phoenix as a primary dependency, providing the underlying capabilities for robust LLM evaluation.

Changed

🌐 API Route Pluralization: Renamed key API routes (e.g., /agent to /agents, /thread to /threads, /event to /events, /suite to /suites, /user to /users) for improved consistency and clarity across the API surface.
🖥️ UI Navigation and Path Updates: Aligned frontend navigation paths and URLs with the new pluralized API routes, ensuring a consistent user experience.
⚡️ Infrastructure Component Upgrades: Updated Docker images for core services, including Phoenix (to v10.0.4), NATS (to 2.11.4), Redis (to 8.0.1), Mongo (to 8.0.9), Llama.cpp, HuggingFace TEI, and Attu, enhancing overall stability and performance.
🛡️ GitHub Actions Permissions: Refined GitHub Actions workflows to use specific PAT_GITHUB_READ_PACKAGES for package access, improving security and clarity in CI/CD processes.
✅ Enhanced Docker Health Checks: Added or improved health checks for several Docker services (Phoenix, Mongo, Llama.cpp, HF-TEI, Attu) to ensure better service readiness and reliability.
💾 Persistent Data Volumes for Databases: Configured data volumes for Redis and MongoDB in Docker Compose files, ensuring data persistence across container restarts.
🎨 UI Theming Consistency: Updated the primary color scheme in the UI theme to leverage surface colors, promoting a more cohesive and modern visual design.
📈 Improved Dashboard Layout Responsiveness: Adjusted grid column layouts in the dashboard for enhanced responsiveness on a wider range of screen sizes.
🔄 Asynchronous Context Handling in Playground: Applied nest_asyncio in the API playground to better manage asynchronous contexts, improving the local development experience.
📝 Updated Localization Strings: Revised several localization strings across German, French, and Italian translations for improved clarity and consistency, including a change from "Rounds" to "Turns" for thread interactions.

Refactor

🧹 Centralized Phoenix Configuration: Relocated the PhoenixConfig module from aihub_agent to aihub_lib, centralizing infrastructure-related configurations for better maintainability.
♻️ Improved i18n Object Loading: Refactored the LocaleHandler's t_object method to dynamically use configured load paths, making i18n resource loading more flexible and robust.
🌱 LlamaIndex Schema Alignment: Updated the VectorPrevNextPostProcessor tests to replace Document with TextNode, aligning with recent schema changes in the LlamaIndex library.

[v0.171.0] - 2025-05-27 - Granular Network Control and Performance Boosts

Added

✨ Introduced Storage Blob Data Contributor Role: A new Azure built-in role is now available, enabling fine-grained permissions for blob storage.
🗺️ Defined Dedicated Subnet CIDRs: Explicit CIDR blocks for Dagster, NATS, Phoenix, WebUI, and core data stores (Cosmos DB, Search, PostgreSQL) have been added, improving network segmentation.
🔑 Granted Storage Blob Access to Dagster: The Dagster identity is now explicitly assigned the Storage Blob Data Contributor role, ensuring proper access to its data lake.

Changed

🚀 Increased Resource Allocations:
- The Dagster webserver now benefits from increased CPU (from 0.5 to 1.5 cores) and memory (from 1Gi to 3Gi) for enhanced performance.
- The PostgreSQL database has been upgraded with larger storage capacity (from 32GB to 64GB) and a more powerful SKU (Standard_B1ms to Standard_B2s) to support demanding workloads.
🌐 Refined Network Provider Initialization: Network providers now receive more comprehensive location details for improved consistency and accuracy in resource deployment.

Refactor

🧹 Decentralized Subnet and NSG Management: The logic for creating subnets and Network Security Groups (NSGs) has been moved from the central Network module into the respective application modules (Dagster, NATS, Phoenix, Stores, WebUI). This change promotes more localized and application-specific network configurations, making the overall infrastructure more modular and easier to manage.
⬆️ Pulumi Azure Native Provider Updates: Updated internal Pulumi provider imports for Cosmos DB (now using cosmosdb instead of documentdb) and Private DNS zones (now using privatedns instead of network), aligning with the latest provider structure.
⚙️ Improved Resource Dependency Handling: Enhanced the provisioning logic for storage accounts and private endpoints to ensure robust and correct dependency resolution during deployment.
🔄 Standardized NATS File Share Naming: Corrected the file share name used for NATS to align with its intended purpose, improving clarity and maintainability.

Fixed

🐛 Resolved Storage Account Name Resolution: Addressed an issue in file share creation where the storage account name might not be fully resolved, ensuring reliable provisioning.

[v0.170.0] - 2025-05-26 - Infrastructure Tooling Enhancements

Added

🚀 New mirror-image Makefile Target: Introduced a utility in aihub_iac to easily pull, retag, and push Docker images between registries, streamlining image management and distribution for internal consumption.

[v0.169.0] - 2025-05-26 - Enhanced DataLakeFile Handling

Fixed

🐛 Improved DataLakeFile Owner Detection: Enhanced the logic for determining the owner of a DataLakeFile to be more robust across different environments. The system now checks the USER environment variable (Unix), USERNAME environment variable (Windows), and falls back to a default pipeline-user if no specific user is found, ensuring proper ownership assignment in varied execution contexts.

Changed

🔄 NATS Container IP Address: Assigned a fixed private IP address to the NATS container instance in Azure deployments. This enhancement improves network stability and predictability for NATS within the infrastructure.

[v0.167.0] - 2025-05-23 - Knowledge Base and RAG System Evolution

Added

✨ Knowledge Base Management UI: Introduced a comprehensive new section in the Admin UI for managing and viewing ingested documents, their individual nodes (chunks), and generated summaries, providing deep insights into RAG data.
📄 Standardized Ingested Data Models: Implemented new Pydantic models (IngestedDocument, IngestedNode) to provide a consistent and rich structure for all ingested data and its metadata, improving data quality and UI display capabilities.
🚀 Milvus Vector Store Integration: Added explicit support and resources for Milvus as a vector store option within the ingestion pipelines, enhancing flexibility for vector database choices.
🤖 Agent Card Component: Introduced a new UI component for displaying agents in a card format, improving the visual presentation and navigation of available agents.
🖼️ Dedicated Markdown Renderer: Added a new UI component for rendering markdown content, enhancing readability and consistency across the application, especially for LLM responses and document nodes.
⚡️ Open WebUI Actions: Implemented new actions (Sources and Tracing) for Open WebUI, allowing users to directly navigate from Open WebUI conversations to detailed source documents and event traces in the AI-Hub Suite.
🛠️ attu Service: Included the attu service in the Milvus Docker Compose setup, providing a web-based GUI for Milvus for easier management and debugging.
🧩 Pipeline Metadata Operations: Added new pipeline operations to ensure default metadata compatibility for RefDoc documents and TextNode objects, standardizing data during ingestion.

Changed

🔄 Core RAG Data Model: Renamed and restructured internal data types from generic Document to more granular IngestedNode across RAG agents, retriever, and reranker events, providing more precise information about retrieved content.
📖 Context Prompt Format: Updated the RAG agent's context prompt format from <DOCUMENT [metadata]> to <REFERENCE_DOCUMENT [metadata]> for clearer demarcation of contextual information from retrieved sources.
📈 Enriched Document & Node Metadata: Expanded the metadata stored with documents and nodes to include fields like number_of_pages, document_title, language, and more detailed timestamp information, improving data richness for UI and analytics.
⚙️ Configurable Document Store Connection: Separated and standardized the CosmosDB connection for the document store, allowing for more robust configuration and isolation from other database connections.
🔑 API Token Controller Permissions: Adjusted permissions for the API Token Controller, making it accessible to a wider range of users instead of being admin-only.
🔗 Data Lake Authentication: Added support for authentication towards Azure Data Lake using an account key, providing an alternative to implicit Azure login for pipeline operations.
🌐 Localization Updates: Comprehensive updates to French, German, and Italian translations to reflect new features and terminology across the application.
📊 Agent & Thread List Enhancements: Implemented selection and visual highlighting for agents and threads in their respective lists within the Admin UI, improving user navigation and context.
⚡️ Azure Document Intelligence API Version: Updated the default API version for Azure Document Intelligence to 2024-11-30.

Fixed

🐛 Event ID Uniqueness in Distribution: Ensured that distributed events to multiple agents in a thread receive unique event IDs, preventing potential tracing inaccuracies and improving the reliability of event tracking.
📏 Metrics Accuracy in Event Summaries: Corrected the aggregation logic for run summaries in PersistedEventEntity to accurately count StartEvents and control_events, ensuring more precise event statistics.

Refactor

🧹 Unified MongoDB Connection Handling: Consolidated MongoDB connection logic across the API, Agent, and Lib components to consistently use ApiConfig().DB_NAME and CosmosDocstoreAccess, simplifying database access management.
♻️ Modular Pipeline Resources: Reorganized pipeline resource factories into more modular functions (e.g., mongo_document_store_resource, milvus_vector_store_resource) to improve clarity and reusability.
🗑️ Removed Document Class: Deprecated and removed the generic Document data model in aihub_lib/nats/events/semantic/retriever, standardizing on the new IngestedNode for all retrieval and reranking contexts.
⚙️ UI Component Specificity: Refined CSS class names and component structure in the web UI for better styling consistency and maintainability.
🗑️ Removed RAG Agent Trigger Script: Removed an outdated trigger script for the RAG agent in the playground environment.

[v0.166.0] - 2025-05-20 - Enhanced RAG Pipelines and System Reliability

Added

🦾 Introduced Dedicated Summary Nodes Processing: A new summary_nodes_factory asset has been added to provide a dedicated, modular pipeline for generating, embedding, and storing summary nodes, enhancing the flexibility of RAG workflows.
⚡️ Convenience Target for Dagster Playground: A new playground target was added to the Makefile, simplifying the setup and execution of the Dagster development environment.

Changed

🔄 Refactored Node Generation Workflow: The primary nodes_factory asset no longer directly creates summary nodes, instead delegating this specialized task to the new summary_nodes_factory for improved modularity and clarity.
⚙️ Updated Playground Configurations: Adjusted default directory, namespace, and vector/document store names within the playground environment for better organization and consistency.

Fixed

🐛 Improved Azure Data Lake Authentication: Corrected an issue where DataLakeAccess was not always passing credentials to AzureBlobFileSystem, ensuring more reliable access to Azure Data Lake resources.
🐞 Ensured Document ID Filterability in Azure AI Search: Fixed an oversight by guaranteeing that the DOCUMENT_ID field is always included in filterable metadata for Azure AI Search vector stores, enabling consistent document-specific queries.
📈 Enhanced Vector Store Query Reliability: Implemented a retry mechanism within VectorStoreIOManager when fetching nodes, making the pipeline more robust against transient issues and improving data retrieval stability.

[v0.165.0] - 2025-05-16 - Enhanced User Experience with Dashboards and Revamped Identity Management

Added

✨ Introduced Customizable User Dashboards: Users can now personalize their home screen with interactive widgets (number, line chart, and bar chart) to visualize key event timeseries data and track agent performance.
🚀 New UserEntity for Comprehensive User Profiles: A dedicated persistence layer has been added to store detailed user information, including names, emails, roles, profile images, and favorite modules, ensuring consistent data across the platform.
🔑 Centralized Azure Graph Service: A new service (AzureGraphService) is introduced to streamline asynchronous interactions with Microsoft Graph API, providing robust and cached retrieval of user profiles, images, and app-specific roles for improved performance and reliability.
📄 New Localization Strings: Added translations for new dashboard, chart, and time range terminology to support multi-language environments.

Changed

🔄 Revamped User Identity Retrieval: The system now prioritizes fetching user details from the new UserEntity for faster access, falling back to external identity providers (like Azure AD) and updating the UserEntity as needed, ensuring user data is always fresh and complete.
⚡️ Optimized Asynchronous API Operations: Numerous API service and controller methods across agent, OpenAI, and thread management functionalities have been updated to use the async/await pattern, significantly improving application responsiveness and resource utilization.
🤖 Flexible API Token Management: API tokens no longer directly store user roles, enhancing security and allowing roles to be dynamically retrieved from the user's primary identity source during authentication.
🔒 Improved OpenWebUI Authentication Integration: The OpenWebuiAuthHandler now leverages the new AzureGraphService to fetch richer user data, including roles, directly from Azure AD, and updated its hash validation mechanism for enhanced security.
🧩 Extensible Multi-Authentication Handler: The TokenAndOauth2Handler now supports a list of authentication handlers for both Bearer and OAuth2 types, allowing for more flexible and layered authentication strategies.
⚡️ Enhanced UI Responsiveness for Thread Updates: The frontend now invalidates relevant queries for thread lists and details upon StopEvent or ExceptionEvent, ensuring immediate reflection of thread status changes in the user interface.
📊 Dynamic Event Timeseries Filtering: Event timeseries queries in the frontend now support optional filtering by agentClass and agentId, enabling more granular insights on dashboard widgets and analytics views.
📄 Updated CLI Token Generation: The generate_api_token.py script has been updated to use Azure CLI for fetching current user information and integrates with the new token creation service, aligning with the revamped token management.

Refactor

🧹 Unified Internal User Representation: Introduced a new UserIdentity object to serve as the canonical internal representation of a user, decoupling it from external DTOs and internal API token structures for cleaner code and better maintainability.
🗑️ Centralized Token User Data: Removed the embedded ApiUser document and roles field directly from the BearerToken entity, streamlining token data to only store the user_oid and relying on the UserEntity for comprehensive user details.

Removed

🗑️ Direct User Details in Bearer Tokens: The practice of embedding full user details and roles within BearerToken documents has been removed, enhancing data normalization and security by centralizing user information in the new UserEntity.

[v0.164.0] - 2025-05-15 - Dependency and Infrastructure Alignment

Refactor

🔄 Refined Infrastructure Type Definitions: Updated environment variable argument types in the Azure Container Instance module for improved consistency and clarity in infrastructure-as-code deployments.

[v0.163.0] - 2025-05-15 - Enhanced Agent Configuration and Infrastructure Tools

Added

✨ Flexible Agent Environment Variables: Introduced support for defining additional_env_vars in agent configurations within the Infrastructure as Code (IAC), allowing users to pass custom environment variables, including secret references, to agent containers for greater operational control.
🛠️ Improved IAC Development Experience: Integrated new development dependencies such as pytest, ruff, mypy, and isort into the aihub_iac project to enhance code quality, testing, and overall developer workflow.

Refactor

⚙️ IAC Resource Naming Refinement: Updated the internal logic for generating AI search resource names within the IAC module for improved consistency.

[v0.162.0] - 2025-05-14 - Enhanced Azure Bot Setup for Improved Integration

Added

⚙️ Azure Bot Setup: The setup_azure_bot.py script now automatically creates a service principal for Azure AD app registrations, streamlining the setup process and ensuring smoother integration with Azure resources.

[v0.161.0] - 2025-05-13 - Streamlined Admin Experience & Data Lake Optimization

Added

✨ Agent Listing Component: Introduced a new component to display available agents in a sortable table, including their name, description, conversational status, and online status.
✨ Display List Component: Added a new component for listing thread displays, offering statistics such as start time, duration, event count, and overall status.
🏗️ Column Layout Component: Implemented a reusable Structural/Column component to provide a consistent visual structure for content columns across the application, including loading indicators and titles.
🖥️ Screen Layout Component: Introduced a new Structural/Screen component to standardize the overall screen layout and background for administrative pages.
📄 Thread Details Component: Added a dedicated component to consolidate and present comprehensive thread details, including integrated display selection and event lists.
🚀 Infinite Thread Scrolling: Introduced a new composable for managing infinite scrolling of threads, complementing the existing standard pagination logic.
🌐 New Translations: Added new internationalization strings for agent lists and thread display details across all supported languages (German, English, French, Italian).

Changed

⚙️ Refined Admin Interface Navigation: Overhauled navigation and layout for Agent and Thread administration pages, leveraging new structural components and a SelectButton for sub-navigation, enhancing consistency and user experience.
🔗 Updated Thread Display URLs: Modified the URL structure for navigating to specific thread displays, improving consistency and clarity (e.g., from /events to /display/[display_id]).
🎨 Improved Service Selection UI: Enhanced the visual design and layout of the service selection popover for a more polished appearance and better usability.
💅 Minor UI Adjustments: Applied various small visual tweaks to component layouts, including panel headers, text sizes, and item alignment for improved aesthetics across the application.
🔄 Switched Thread Display Component: Replaced the direct use of EventList with the new ThreadDetails component for OpenAI service thread views, ensuring consistent feature presentation.
📈 Timeseries Chart Library: Switched to VueApexChart for event timeseries visualizations, providing enhanced charting capabilities.

Refactor

⚡️ Optimized Data Lake Access: Refactored DataLakeAccess to cache the AzureBlobFileSystem client, reducing redundant object creation and improving performance when interacting with Azure Data Lake Storage.
🧹 Streamlined Event List Component: Simplified the EventList component by externalizing display selection logic into a parent component, making it more focused and reusable.
🔄 Thread List Pagination Refactor: Reworked the useThreads composable to use standard pagination instead of infinite scrolling for the main thread list, improving control and user experience for large datasets.
🗺️ Dynamic Breadcrumb Generation: Refactored the default layout to dynamically generate breadcrumbs based on the current route path, providing more accurate and flexible navigation.
🧹 Standardized I18n Keys: Performed a significant cleanup and restructuring of internationalization keys across all language files for better consistency and maintainability.
📄 I18n Usage Consistency: Standardized the way internationalization functions are called (from $t() to t()) throughout numerous web components and pages for improved consistency.

[v0.160.0] - 2025-05-13 - Refined Embedding Model Configuration

Changed

🔄 Enhanced Azure OpenAI Embedding Parameter Handling: The dimensions parameter for Azure OpenAI embeddings now robustly supports the NotGiven type, enabling more precise control over API requests and aligning better with the underlying OpenAI library's parameter conventions.

Refactor

🧹 Streamlined Embedding Model Configuration: The dimensions parameter has been moved from the generic embedding model configuration to specific provider configurations, improving modularity and ensuring parameters are only present where relevant.

[v0.159.0] - 2025-05-13 - Enhanced LLM Embeddings and Event Query Performance

Added

✨ Enhanced Embedding LLM Configuration: Introduced support for specifying the dimensions parameter in embedding LLM configurations, allowing for more control over the output vector size, particularly for models like text-embedding-3 and later.
⚡️ Improved Event Persistence Querying: Added a new index on event_data.created_at for PersistedEventEntity, significantly enhancing the performance of time-based queries on persisted event data.

[v0.158.0] - 2025-05-12 - Core API Simplifications

Refactor

🧹 Streamlined Pydantic Model Creation: Removed redundant __base__=BaseModel arguments from create_model calls in the API event model generation, simplifying the code while maintaining full functionality.

[v0.157.0] - 2025-05-08 - Infrastructure Enhancements for WebUI Deployment

Refactor

🧹 Refined WebUI API Key Configuration: Explicitly set optional API key fields to None by default in OpenWebUIConfig for improved clarity in infrastructure deployments.
⚙️ Standardized WebUI Container Naming: Renamed an internal method to container_app_name within WebUIConfig for consistent resource naming conventions.
🔗 Integrated Registry Settings for WebUI: Added RegistrySettings to WebUIConfig, allowing for more robust and configurable container registry management for WebUI deployments.

[v0.156.0] - 2025-05-08 - Introducing Comprehensive Event Statistics & UI Overhaul

Added

✨ Event Timeseries API: Introduced a new API endpoint (/event/timeseries/{time_range}) to retrieve time-based statistics for events, allowing for detailed analysis of event counts over various periods (1h, 24h, 30d, 365d) with adaptive resolution.
📊 Dynamic Event Charts: Implemented new UI components and composables (EventStatistics, EventTimeseries) to visualize event data, including agent invocations, end events (success, human/bot/agent in the loop, errors), and delegations, with interactive time range selection.
📈 Core Timeseries Aggregation: Added robust MongoDB aggregation logic within aihub_lib to efficiently calculate event counts across defined time buckets, supporting the new analytics capabilities.
🗺️ Localization for Analytics: Extended internationalization support to include new translation keys for event statistics chart titles and metrics (costs, duration).
🎨 Charting Library Integration: Incorporated ApexCharts into the web UI, enabling rich and interactive data visualizations.

Changed

💅 Improved Chat Message Styling: Enhanced the visual presentation of user chat messages for better readability and distinction.
🚀 Overhauled Agent & Thread Overview Pages: Redesigned the agent and thread overview pages to prominently feature new event statistics, duration, and cost metrics, replacing older status indicators with more comprehensive data.
⏱️ Adjusted Thread Data Stale Time: Increased the cache stale time for thread data fetching to 5 minutes, improving performance by reducing redundant API calls for frequently accessed thread information.
🔄 Refined Thread Event Updates: Optimized thread event fetching and added cache invalidation on StopEvent or ExceptionEvent to ensure more immediate updates for thread status and statistics.
🗓️ Standardized Date Handling: Improved date formatting and ensured consistent UTC ISO format with timezone awareness for all thread-related timestamps in API responses.
↔️ Enhanced Thread Panel UX: Adjusted the positioning and styling of the thread panel close button for improved user experience and accessibility.

Fixed

🐛 Corrected Event Aggregation Logic: Resolved an issue in PersistedEventEntity's aggregation pipeline to ensure accurate counting of events by de-duplicating entries, leading to more reliable run statistics.

Refactor

🧹 Terminology Standardization: Renamed "latency" to "duration" across the API DTOs, database models, and UI components for clearer and more consistent terminology.
⚙️ Centralized Thread Access Checks: Refactored user access validation for threads into a dedicated utility method within ThreadService, promoting code reusability and maintainability.
🔄 Updated Query Naming Conventions: Standardized query keys and transitioned from isLoading to isPending in various Vue composables for consistency with pinia/colada's query state management.

[v0.155.0] - 2025-05-08 - Enhanced Speech-to-Text Capabilities and Infrastructure Updates

Added

🚀 Enhanced Speech-to-Text (STT): Introduced intelligent audio chunking to seamlessly process and transcribe large audio files by splitting them at silence points and merging the results, overcoming typical API size limitations. This significantly improves the reliability and reach of STT services.
📦 FFmpeg Requirement: Integrated the necessary pydub library and FFmpeg installation steps within aihub_api and the CI/CD pipeline to support advanced audio processing functionalities.
💾 Persistent MongoDB Data: Ensured data durability for MongoDB by adding a persistent volume to the docker-compose.yml configuration, preventing data loss across container restarts.
🧪 Comprehensive Audio Tests: Added extensive test coverage for the new audio chunking and transcription merging functionalities to guarantee robust performance and accuracy.

[v0.154.0] - 2025-05-08 - Streamlined Versioning and IAC Integration

Changed

🔄 Automated Versioning Scope: The aihub_iac project has been fully integrated into the automated version tagging workflow, ensuring consistent release management and version alignment across all core components.

[v0.153.0] - 2025-05-08 - Internal Tooling and CI/CD Scope Expansion

Changed

⚙️ Expanded Semantic PR Scopes: Added iac (Infrastructure as Code) and ci-cd as recognized scopes for semantic pull requests, enhancing consistency and automation for relevant contributions.

[v0.152.0] - 2025-05-08 - Introducing Comprehensive Azure Infrastructure as Code

Added

✨ Comprehensive Azure IaC Module: Introduced aihub_iac, a new Pulumi-based Infrastructure as Code module for deploying the entire AIHub platform on Azure. This enables consistent, automated, and version-controlled provisioning of all necessary cloud resources.
🚀 Modular and Secure Infrastructure Deployment: The new IaC module features a component-based design with dedicated providers for network, identity, and resource creation, ensuring modularity, reusability, and secure private networking for core services.
🌐 Automated Microservice Deployment: Integrated Pulumi resources for automated deployment of all AIHub microservices, including API, Bot, and Phoenix (as Azure App Services) and Agents, NATS, Redis, Dagster, and OpenWebUI (as Azure Container Instances or Container Apps).
💾 Managed Data Stores: Provisioned and configured managed data services, including Azure Cognitive Search for vector capabilities, Azure Cosmos DB for document and API storage, and Azure PostgreSQL Flexible Server with pgvector extension for relational and vector data.
⚙️ Core Infrastructure Services: Established foundational infrastructure components such as a private Virtual Network with segmented subnets, Network Security Groups, and Private Endpoints to enhance security and isolation.
📄 Expanded On-Premises Technology Overview: Enhanced the README.md to reflect the newly supported and integrated on-premises technologies, specifically detailing Postgres (SQL DB), PG-vector for vector capabilities, and Docker for application environments.

[v0.151.0] - 2025-05-06 - Comprehensive Thread Management & Enhanced Agent Insights

Added

✨ Comprehensive Thread Statistics: Introduced detailed metrics for each thread, including event counts, conversation turns, pending operations, error states, total LLM costs, and interaction latency. These statistics are aggregated per display and individual runs for granular insights.
🦾 Agent Persistence: Agents are now persisted in the database, allowing for discovery and retrieval of both online and offline agents and their configurations.
📄 Paginated Thread Views: New API endpoints and UI sections to retrieve paginated lists of threads associated with users and specific agents, improving navigability for large datasets.
🔄 Thread History Reconstruction: Enabled the API to reconstruct full chat message history for OpenAI API calls based on persisted thread events, facilitating seamless conversation continuity.
🗺️ Workflow Hierarchy Visualization: Introduced a new visual component to display the hierarchical structure of thread executions, illustrating how displays and runs are nested.
🌐 Locale-Aware Event Display: Implemented internationalization for event names and descriptions in the UI, ensuring that event details are displayed in the user's preferred language.
💬 Multimodal Chat Message Support: Extended the backend and frontend to properly handle and display chat messages containing various content types, including images and audio.
🛡️ GuardEvent for Tracing: A new event type (GuardEvent) has been added to provide better tracing and visibility into security and policy guardrail evaluations within agent workflows.
🚦 RouterEvent for Decision Visibility: Introduced RouterEvent to clearly mark and visualize points in the workflow where an LLM makes a decision on the optimal next path.
💥 ExceptionEvent as Semantic Event: The ExceptionEvent now inherits from SemanticEvent, enabling its data to be reported to OpenInference-compatible tracing systems for enhanced error monitoring.
🧪 Playground Testing Agents: Added new test agents in the playground to demonstrate and validate complex workflow patterns, including explicit routing and bot-in-the-loop interactions.
🧩 ControlAndDisplayEvent Base Class: Introduced a new helper base class to simplify the creation of events that serve both as workflow control signals and user-facing display messages.
👤 New UI Components: Added dedicated UI components for displaying Agent and User Avatars, enhancing visual consistency and clarity.

Changed

🧹 Internal i18n Key Refactoring: Standardized internationalization keys across agents from agents.<scope> to agent.<scope> for improved consistency.
⚡️ Optimized Agent Workflow Control: Refactored ExpertGroundedAgent and LLMWrappingAgent to return direct StopEvent or AgentInTheLoop.request where appropriate, streamlining their workflow integration.
🚀 Streamlined API Thread Operations: Refactored thread management operations in aihub_api to directly interact with the persistence layer (MongoDB), removing NATS as an intermediary for these specific tasks, which simplifies the architecture and improves performance.
🌐 Improved WebSocket Locale Handling: The WebSocket connection now explicitly extracts the user's locale from request headers for more accurate language presentation of real-time events.
🧰 Updated UI Components: Enhanced existing UI components, including ChatMessage, CostsTable, and event display components, to integrate with new data structures and internationalization.
📦 Dependency Updates: Updated numerous core dependencies, including OpenTelemetry, LlamaIndex, and OpenAI libraries, for improved performance, stability, and new features.
🧹 Refactored Frontend Data Management: Migrated existing Pinia stores to use @pinia/colada queries and standard Nuxt composables for a more modern and efficient data fetching and state management pattern.
📈 Enhanced Logging: Replaced numerous logger.error calls with logger.exception across the codebase to ensure full stack trace capture for better debugging.

Fixed

🐛 Robust Event Deserialization: Improved the BaseEvent deserialization logic to more accurately handle events with multiple inheritance paths, preventing potential data interpretation issues.
🪲 AnyUrl Serialization in Events: Addressed an issue where AnyUrl fields within chat messages were not correctly serialized in UserMessageEvent, ensuring proper display of embedded URLs.

Removed

🗑️ Deprecated Thread & Agent UI Pages: Replaced outdated thread and agent management pages with new, more robust, and feature-rich implementations.
🗑️ Redundant Playground Components: Removed legacy playground components like UserMessageEvent form and Thread/Chat that are superseded by new UI and testing features.

[v0.150.0] - 2025-05-04 - LLM Evolution: Introducing Gemini and Intelligent Content Summarization

Added

✨ Google Gemini LLM Support: Introduced GeminiLLMConfig and integrated Google Gemini models into the aihub_api and aihub_lib, allowing for more diverse LLM choices.
📄 Recursive Document Summarization: Added a new RecursiveNodeSummarizer and integrated it into the RAG pipeline (aihub_pipeline), enabling hierarchical summarization of documents to improve retrieval and context understanding.
🚀 New Pipeline Playground: A comprehensive Dagster pipeline playground (aihub_pipeline/playground) was introduced to facilitate testing and development of RAG ingestion workflows with new features like recursive summarization.

Changed

🦾 Stricter Agent Description Guard Prompt: Updated the prompt for the agent description guard across all languages to enforce a stricter JSON output format, ensuring more reliable structured responses.
⚡️ Improved Recursive Summarization Parameters: Adjusted the minimum summarization length and enhanced locale handling within the RecursiveNodeSummarizer for more flexible and accurate document summaries.

Refactor

🔄 Standardized OpenAI-like LLM Naming: Renamed SelfHostedLLMConfig and SelfHostedLLMParameter to OpenaiLikeLLMConfig and OpenaiLikeLLMParameter across the codebase. This standardizes the configuration for models that adhere to the OpenAI API specification but are hosted locally or externally.
🧹 Internal Module Reorganization: Reorganized internal imports and module structures within aihub_agent and aihub_bot to improve code clarity and maintainability.

[v0.149.0] - 2025-05-04 - Enhanced LLM Flexibility with Gemini Support

Added

✨ Google Gemini LLM Support: Introduced GeminiLLMConfig and associated configurations, enabling seamless integration and usage of Google Gemini chat models via their OpenAI-compatible API.

Refactor

🧹 Standardized OpenAI-like LLM Configuration: Renamed SelfHostedLLMConfig to OpenaiLikeLLMConfig to more accurately reflect its compatibility with various OpenAI-like API endpoints, improving clarity and broad applicability across the codebase.
🔄 Updated RAG Agent Module Path: The RAGAgent and its related configuration files have been reorganized and moved to a new, more logical module path (aihub_agent.agents.RagAgent), streamlining the internal structure of agent implementations.

[v0.148.0] - 2025-04-24 - Enhanced RAG Agent Context Sufficiency

Changed

🦾 Refined RAG Agent's context handling: The RAG Agent now provides more precise information to its context sufficiency guard regarding the availability of further retrieval attempts (hops), allowing for more informed decision-making.
⚙️ Improved Context Sufficiency Guard: The guard's response structure and internal logic now adapt dynamically based on whether more retrieval hops are available. This leads to more accurate reasoning and a tailored output, providing clearer indications when the maximum retrieval attempts have been reached or when further queries are still possible.

[v0.147.0] - 2025-04-24 - Enhanced Azure AI Search Vector Store Configuration

Added

✨ Configurable Azure AI Search Embedding Dimensions: Introduced a new dimensions parameter for Azure AI Search vector stores, allowing users to specify the embedding dimensionality (defaulting to 3072). This feature ensures better compatibility and performance by enabling the vector store to precisely match the output dimensions of different embedding models used for RAG operations. This enhancement is available through aihub_lib and integrated into aihub_pipeline resources and definitions for seamless configuration.

[v0.146.0] - 2025-04-17 - Enhanced Conversation Management and Bot Persistence

Added

✨ Introduced Conversation Tracking: Added a new ConversationTracker entity to distinguish between conversations that automatically expire due to inactivity and those explicitly deleted by users (e.g., in Microsoft Teams).
💾 Configurable Conversation Time-To-Live (TTL): Implemented MongoDB TTL indexing for ConversationEntity based on conversation inactivity, allowing for automatic cleanup of old conversations.
🗣️ Expired Conversation Notification: Bots now inform users when they resume an expired conversation, providing clarity that previous messages are no longer available.
⚙️ Customizable Typing Indicator Timeout: Added a typing_timeout_seconds parameter, allowing configuration of how long the bot's "typing" indicator is shown before it times out.

Changed

⏱️ Conversation Inactivity Tracking: The ConversationEntity now automatically updates a last_activity timestamp with every user message, crucial for the new TTL mechanism.
🚀 Bot Runner Configuration: The BotRunner now accepts a conversation_ttl_days parameter to easily configure conversation persistence across all mounted bots.
💬 Bot Route Configuration: Agent and OpenAI chat controllers now allow specifying typing_timeout_seconds for fine-grained control over the typing indicator behavior for specific chat endpoints.
🧹 Conversation Reset Logic: Improved handling of conversation resets in platforms like Microsoft Teams by explicitly marking conversations as deleted when a user initiates a new chat session with the same conversation ID.

[v0.145.0] - 2025-04-16 - Streamlined AI Hub Connections and Agent Playgrounds

Added

✨ New Long-Running Agent Example: Introduced a new playground example demonstrating how to create and manage agents that stream output incrementally over an extended period, showcasing real-time progress.

Changed

🔄 Improved AI Hub Connection Handling: The pipe function in webui_pipelines/aihub_connection.py now dispatches to dedicated handlers for streaming and non-streaming chat completion requests, ensuring more robust and explicit management of different response types.
📈 Enhanced Model Fetching Error Reporting: Improved error handling and reporting when fetching available models from the AI Hub API, providing clearer feedback on connection issues.

Refactor

🧹 Asynchronous HTTP Client Migration: Switched the underlying HTTP client in webui_pipelines/aihub_connection.py from requests to httpx.AsyncClient for improved asynchronous performance and reliability in handling API requests.
⚡️ Consistent Source Event Retrieval: Refactored the logic to consistently query and emit source events (retrieved documents) after both streaming and non-streaming chat completions are finished, providing more reliable RAG context in the UI.
⚙️ Stream and Non-Stream Request Specialization: Separated the core logic for handling streaming and non-streaming chat completion requests into distinct, specialized functions (pipe_stream and pipe_non_stream) for better code organization and maintainability.

[v0.144.0] - 2025-04-15 - Introducing Adaptive Context Retrieval for RAG Agents

Added

✨ Introduced Multi-Hop Retrieval for RAG Agents: RAG agents can now perform multiple retrieval attempts (hops) with dynamically generated and refined queries if the initial context is insufficient, significantly improving the chances of finding relevant information.
📄 New ContextInsufficientWithQueryEvent: This event has been added to facilitate the multi-hop retrieval process by carrying a revised query for subsequent retrieval attempts when the initial context is deemed insufficient.

Changed

🔄 Enhanced Context Sufficiency Guard: The context_sufficient_guard now intelligently generates new, distinct queries for further context retrieval. It considers previous query attempts to avoid repetition and guides the LLM to explore higher-level conceptual hierarchies when direct approaches fail.
📈 Increased LLM Context Window: The default context size for self-hosted LLMs (e.g., Llama-3.2-1B-Instruct-GGUF) in docker-compose configurations and tests has been significantly increased from 512 to 8192 tokens, enabling the processing of more extensive context.
⚙️ RAG Agent Configuration for Hops: A max_hops configuration option has been added to RAGAgentConfig, allowing users to control the maximum number of retrieval attempts the RAG agent will perform.
🧪 Updated RAG Agent Playground: The RAG Agent playground now defaults to using Azure OpenAI LLM and includes a new scenario demonstrating the multi-hop retrieval capability.

Refactor

🧹 Refined Guard Result Naming: Renamed the GuardResult class to ContextGuardResult within aihub_lib for improved clarity and semantic consistency.
🗑️ Cleaned Azure OpenAI Embedding Parameters: Removed the dimensions field from AzureOpenAIEmbeddingParameter for schema simplification, as it was not actively used.

[v0.143.0] - 2025-04-11 - Agent Evolution, Advanced Routing, and Integration Enhancements

Added

✨ Expert-Asking Agent: Introduced a new agent that automates posing questions to designated experts via Slack, manages follow-up questions for insufficient answers, and stores sufficient responses as knowledge snippets in OpenWebUI.
🦾 Expert-Grounded Agent: Added a new agent that leverages existing context and conversation history to answer user queries. If knowledge is insufficient, it intelligently prompts the user for consent to consult a human expert via the new Expert-Asking Agent.
🚀 OpenWebUI SDK Integration: Integrated a comprehensive Software Development Kit (SDK) for OpenWebUI within aihub_lib, enabling direct programmatic interaction with OpenWebUI's chat, file management, and knowledge base functionalities.
⚡️ LLM-Based Routing Framework: Implemented a new dynamic routing mechanism that allows agents to use Large Language Models (LLMs) to make intelligent decisions on the next step in a workflow, enhancing agent adaptability and complexity.
📈 RAG Embedding Batch Size Configuration: Added a new configuration option for the RAG agent within docker-compose-webui.yml to specify the batch size for embedding generation, potentially improving performance.
📦 New Dependency: Included the stringcase library to support advanced string manipulation for improved filename generation in agents.

Changed

🌐 Enhanced Internationalization (i18n) Support: Improved locale handling across the API, bot, and pipeline components, allowing user language preferences to be passed consistently to agents for more localized responses.
🔄 LLM API Version Update: Updated the Azure OpenAI LLM API version to 2024-12-01-preview across all configurations and playground examples for improved compatibility and access to the latest features.
🚨 Improved Error Logging: Replaced traceback.print_exc() with logger.exception(e) for more structured and robust error logging within the dispatcher and NATS subscribers, enhancing debugging and system stability.
📄 Displayable Workflow Events: Made key internal workflow events, such as Limit Chat History and Standalone Question Condenser, displayable in the user interface to provide greater transparency into agent processing. The Start Event is also now explicitly displayable.
🛑 Streamed LLM Responses: Enhanced the LLM event displayer to allow streamed Large Language Model responses to directly signal the conclusion of a workflow run, streamlining agent completion.
🔧 Agent Runner Validation: Added robust type and subclass validation for agent types during initialization of the AgentRunner, preventing common misconfigurations.
⚙️ OpenWebUI Docker Compose Configuration: Modified default OpenWebUI settings in docker-compose-webui.yml to disable automatic configuration resets and channels on startup, and enabled API key authentication by default.
📝 Context Event Documentation: Updated the ContextInsufficientEvent and ContextSufficientEvent with clearer docstrings and improved field definitions for better understanding and usage.
💬 Playground BotInTheLoopAgent Update: Updated the playground BotInTheLoopAgent to explicitly include the Slack channel ID when invoking bot interactions.

Refactor

🧹 Agent Base Class Simplification: The core Agent base class has been moved and simplified, removing its previous abstract categorization for a more direct and accessible inheritance model.
📂 Agent Directory Restructuring: Reorganized agent modules and their associated files (e.g., FewShotAgent, LLMWrappingAgent, RagAgent, WebuiAgent) into flatter, more intuitive directory structures for improved discoverability and maintainability.
📦 Centralized Common Events: Relocated frequently used control events, such as LimitChatHistoryEvent and StandaloneQuestionCondenserEvent, from aihub_agent to aihub_lib to centralize common event definitions and promote reuse across components.
🏷️ Prompt Template Naming Consistency: Renamed the user_question variable to user_query within prompt templates for better consistency and clarity.
🔗 Robust Event Serialization: Improved the event serialization and deserialization logic to correctly handle lists containing embedded event or model instances, enhancing the reliability of complex event structures.

Fixed

🐛 Thought Event Formatting: Corrected an issue where ThoughtEvent content might have leading/trailing whitespace in display tests, ensuring consistent formatting.
🛠️ Event Naming Consistency: Rectified an inconsistency in event naming for HumanInTheLoopResponseEvent within the chat service, improving internal consistency.

[v0.142.0] - 2025-04-10 - Streamlined Bot-in-the-Loop Interactions and Improved Azure Bot Setup

Changed

🔄 Bot-in-the-Loop Request API Update: The BotInTheLoop.invoke method now exclusively requires the Slack channel ID (e.g., C08MCK6LEBY) instead of the full conversation ID. The bot automatically derives the complete conversation identifier, simplifying agent configuration.
⚙️ Internal Slack Conversation ID Management: The bot now dynamically retrieves and caches Slack bot and team IDs using the auth.test API, which are then used to construct full conversation identifiers. This enhances flexibility and reduces manual configuration requirements.
🧹 Refined Bot-in-the-Loop Thread Tracking: Internal thread management for Bot-in-the-Loop interactions now utilizes Slack's thread_ts (timestamp) for more robust and accurate tracking of conversations.
🛠️ Improved Azure Bot Setup Script Robustness: The setup_azure_bot.py script now automatically checks for existing Azure AD app registrations, parses command output more robustly, and includes auto-confirmation for deletions, streamlining the Azure Bot deployment process.

Added

🗣️ Responder Information in Bot-in-the-Loop Responses: BotInTheLoop.response events now include an optional responder field, providing details such as user_id and user_name of the Slack user who provided the input, enabling better accountability and context.
📄 New Slack Utilities Module: Introduced SlackUtils to encapsulate logic for interacting with the Slack API, specifically for fetching essential bot and team IDs, supporting the new conversation ID management.

[v0.141.0] - 2025-04-09 - Enhanced Test Infrastructure and Core Library Updates

Changed

🚀 Improved CI/CD Docker Health Checks: Enhanced the GitHub Actions backend test workflow with more robust Docker service health checks. This includes added timeout mechanisms, checks for Docker availability, and clearer diagnostics, leading to greater test reliability and better troubleshooting.
⬆️ Upgraded llama-index-vector-stores-milvus: Updated the core aihub_lib dependency for Milvus vector store integration from version 0.6.0 to 0.7.3, incorporating upstream improvements and increased stability.

Fixed

🐛 Stabilized Milvus Vector Store Tests: Addressed intermittent test failures related to Milvus vector store operations by introducing a small, strategic delay after data ingestion, ensuring that the vector store is fully ready for subsequent test queries.

[v0.140.0] - 2025-04-09 - Enhanced Agent-Bot Interaction with Bot-in-the-Loop Capabilities

Added

🤖 Bot-in-the-Loop (BiTL) Workflows: Introduced a powerful new capability allowing agents to pause execution and solicit direct input or approval from human operators via a bot, currently integrated with Slack.
✨ New BotInTheLoopAgent: A dedicated agent now available in the playground to demonstrate and initiate these interactive Bot-in-the-Loop requests.
💬 BotInTheLoopBot: A specialized bot has been added to handle and relay responses from human operators back into ongoing agent workflows, ensuring seamless interaction.
📬 BiTL Event Framework: Implemented new NATS event types (BotInTheLoopRequestEvent and BotInTheLoopResponseEvent) within aihub_lib to standardize communication for human intervention steps.
🚀 Core System Support for BiTL: Updated the agent dispatcher and external event distributor to fully support the routing and processing of Bot-in-the-Loop events.
⚙️ IntelliJ Run Configuration for BiTL Agent: Added a convenient run configuration for developers to easily run and debug the BotInTheLoopAgent within IntelliJ.

Refactor

🧹 Unified Chat Bot Architecture: Reorganized the aihub_bot's internal structure by consolidating chat-related bots (Agent and OpenAI) into a new chat subdirectory, enhancing code organization and maintainability.
♻️ Routes Service Enhancements: Minor internal refactoring within RoutesService to streamline credential retrieval.

Changed

🛠️ Improved Dependency Management: The switch_dependencies.py script now automatically runs poetry install after poetry lock, simplifying environment setup for developers.

[v0.139.0] - 2025-04-08 - Visualizing Agent Journeys: UI Refresh and Event System Evolution

Added

✨ Enhanced Frontend Event Visualization: Introduced a comprehensive set of new Vue components (AgentInTheLoopRequestEvent, ChunkEvent, EmbeddingEvent, LLMCostEvent, LLMEvent, RerankerEvent, RetrieverEvent, ThoughtEvent, ToolEvent, UserMessageEvent, StopEvent, HumanInTheLoopRequestEvent, HumanInTheLoopResponseEvent, and more) to provide rich, type-safe, and user-friendly displays for various event types in the UI, offering deeper insights into agent behavior.
🚀 Integrated Open WebUI as a Core Frontend: Shipped a full integration of Open WebUI as a powerful chat interface, including dedicated Docker Compose configurations (docker-compose-webui.yml) and pipeline scripts (webui_pipelines/aihub_connection.py, webui_pipelines/suite_connection.py) to connect it to the AI-Hub backend.
📈 New Event Tracing & Source Display: Implemented a new "Sources" drawer in the UI (accessible via Open WebUI integration) that visualizes detailed backend events (like retrieved documents) for any given chat session, significantly improving observability and debugging capabilities.
🦾 Dedicated Frontend Testing Agent: Introduced FrontendTestingAgent and its associated configurations, providing a specialized agent to simplify and accelerate the development and testing of new frontend event display features.
🔑 Azure AD Integration for Open WebUI Users: Enhanced Open WebUI authentication to leverage Azure Active Directory for user and role information, providing more robust security and role-based access control.
⚡️ Performance Improvements for User & Thread Data: Added caching for Azure AD user information and thread-user lookups, reducing database load and improving response times for common operations.
📚 Comprehensive Product Pitch Documentation: Included a new 0_product_pitch.md document along with supporting media, offering a high-level overview of the AI-Hub's vision, architecture, and benefits.
🔌 Standardized WebSocket Dependencies: Introduced new dependency injection utilities (use_ws_manager_ws, use_ws_sender_ws, use_external_event_distributor_ws, use_nats_ws) to streamline the integration of NATS, WebSocket management, and external event distribution in WebSocket contexts.

Changed

🔄 Core Event System Revitalization: Undertook a major refactor of the internal event system across aihub_agent and aihub_lib, standardizing event identification using event_name (string identifier) and event_class (Python type). This improves clarity, consistency, and robustness in event processing and serialization.
🛡️ Refined AuthHandler for WebSocket Authentication: Updated the AuthHandler interface to include a direct authenticate_token method, improving flexibility and consistency for WebSocket connections and other non-HTTP authentication scenarios.
🧹 Unified ID Conversion Utility: Centralized string-to-ObjectId conversion into a new str_to_object_id utility, eliminating redundant logic in ThreadEntity and other modules.
💬 Enhanced Chat Service Context Handling: Improved the chat service's ability to handle and pass thread_id and display_id for better context tracking in conversational flows.
🧠 Structured Reasoning Content in Responses: Modified ChunkEvent to include an optional reasoning_content field and updated the chat service to expose this in responses, allowing for streaming of internal agent thoughts directly to the frontend.
📊 Granular Event Filtering for API: Expanded the GET /event/ API endpoint to support filtering by thread_id, display_id, and event_class, enabling more precise retrieval of historical events.
📦 API Payload Structure for WS Events: Reworked the WSServerEvent payload to use a polymorphic event field instead of a generic event_data dictionary, providing strong typing and easier consumption for frontend clients.
🐛 Fixed Duplicate Agent Discovery Entries: Addressed an issue where agent discovery could lead to duplicate agent entries in the API response.
📃 Updated Documentation Examples: Aligned code examples in aihub_doc/5_agents_in_detail.md to reflect the new event.event_name property usage.
🚀 Improved Frontend Serving: Enhanced the Runner.mount_frontend method to better serve Nuxt.js SPAs, including specific handling for _nuxt and _fonts static assets and a catch-all route for client-side routing.
⚙️ Development Environment Updates: Updated CLIENT_ID in development environment configurations and adjusted OpenAI base URLs.

Removed

🗑️ Deprecated Open WebUI Page: Removed the previous aihub_web/pages/module/webui/index.vue page, replaced by the new, more robust integration.
🗑️ Redundant ThreadEntity.to_thread_id: Eliminated the now-obsolete to_thread_id method from ThreadEntity following the introduction of a centralized ID conversion utility.
🗑️ Consolidated ExternalEvent Export: Removed a redundant __init__.py file related to ExternalEvent as its definition is now centrally managed.
🗑️ Outdated WebSocket Receiver Modules: Cleaned up and removed old aihub_api/aihub_api/sockets/receiver modules, replaced by the new, refactored WebSocket architecture.

[v0.138.0] - 2025-04-07 - Core Infrastructure and Testing Improvements

Changed

🚀 Upgraded Milvus Vector Store Integration: Updated the llama-index-vector-stores-milvus dependency from 0.3.0 to 0.6.0, bringing the latest enhancements and fixes for vector store operations.
🧪 Enhanced Asynchronous Test Stability: Introduced session-scoped asyncio event loops within test fixtures to ensure more reliable and consistent execution of asynchronous tests.
⚙️ Streamlined Local Development Setup: Refined the NATS service configuration in docker-compose.yml by removing the health check, which helps optimize the local development environment startup.

[v0.137.0] - 2025-04-07 - Core Event System Enhancements

Changed

🔄 Improved Human-in-the-Loop (HITL) Event Handling: Enhanced the serialization and deserialization of HumanInTheLoopResponseEvent within the chat service to ensure proper type information is consistently included, leading to more robust and reliable event processing.

[v0.136.0] - 2025-04-02 - Improved LLM Guarding and Structured Output

Refactor

🧹 Refactored ContextSufficientGuard: The internal implementation of the context_sufficient_guard has been updated to leverage llm.structured_predict for more robust and localized structured output parsing, enhancing the reliability of context sufficiency checks.
⚙️ Modernized LLM Interaction: Switched from LLMTextCompletionProgram to direct llm.structured_predict calls, improving how structured data is extracted from LLM responses and enabling dynamic Pydantic model generation for better internationalization.

Changed

📄 Updated Internationalization Definitions: Adjusted translation files for context_sufficient_guard to support the new structured output fields and streamline prompt templates, improving clarity and consistency across languages.

Refactor

🧹 Unified Bot Logic with BaseChatBot: Introduced a new base class, BaseChatBot, to centralize and standardize common functionalities across all chatbot types, including message processing, typing indicators, and conversation management.
🔄 Redesigned Completion Handling: Refactored the core Service into a more abstract CompletionHandler to implement a strategy pattern, streamlining how different AI models (e.g., Agents, OpenAI) generate responses.
🚚 Restructured Bot Modules: Reorganized bot-related files and modules (e.g., ContentExtractor, AgentChatService, OpenaiChatService) into a clearer, more maintainable directory structure.
🧹 Streamlined Bot Implementations: Simplified AgentChatBot, StreamAgentChatBot, OpenaiChatBot, and StreamOpenaiChatBot classes by migrating shared logic to the new BaseChatBot.
⚙️ Standardized Database Naming: Aligned database and collection names for bot paths (aihub_bot to aihub, paths to bot_paths) for consistency.

Added

🌍 Internationalization for Bot Error Messages: Implemented support for localized error messages, providing users with clearer and more relevant feedback in their preferred language.

Fixed

🐛 Improved OpenAI API Error Handling: Enhanced the OpenaiCompletionHandler to specifically catch and display detailed error messages from OpenAI's API, offering better transparency for API-related issues.

Changed

📝 Flexible Locale Handling: Updated LocaleHandler.t_object to allow optional locale arguments, providing more flexible usage when retrieving translated objects.

[v0.134.0] - 2025-04-01 - Core Component Synchronization

Changed

🔄 Synchronized Component Versions: All core microservices (aihub_agent, aihub_api, aihub_bot, aihub_pipeline) and the shared aihub_lib have been updated to v0.134.0, ensuring consistent internal dependency alignment across the platform.
⚙️ Makefile Tag Update: The Makefile's default TAG for remote core operations has been updated to v0.134.0 to reflect the new coordinated release version.

[v0.133.0] - 2025-04-01 - Improved API Error Handling with Custom Status Codes

Added

✨ ExceptionEvent Custom Status Code: Introduced a new http_status_code field to the ExceptionEvent to allow services to specify precise HTTP status codes for exceptions.

Changed

🚀 API Error Response Precision: The API now leverages the http_status_code from ExceptionEvent to return more accurate and specific HTTP status codes to clients, enhancing error visibility beyond a generic 500 internal server error.
🔄 Agent Service Error Propagation: The AgentService has been updated to explicitly allow the return of an ExceptionEvent, ensuring more detailed error information is propagated from agent operations.

[v0.132.0] - 2025-04-01 - Improved Chatbot Reliability and User Feedback

Changed

🔄 Enhanced Chatbot Error Handling: Standardized and centralized the handling of unexpected errors across all chatbot types (Agent and OpenAI, streaming and non-streaming) to provide more consistent user feedback and better internal logging.
⚡️ Introduced Response Timeout for Chatbots: Chatbots now detect and notify users if a response takes an unusually long time, preventing indefinite waiting and indicating potential issues.
🔗 Improved Internal Exception Propagation: Enhanced the system's ability to propagate specific ExceptionEvent types from core services, allowing for more precise and robust error management within the chatbot framework.

[v0.131.0] - 2025-04-01 - Enhanced Markdown Parsing Accuracy

Fixed

🐛 Markdown Parsing: Improved the accuracy of Markdown document parsing by automatically unescaping HTML entities, ensuring that content extracted from Markdown files is correctly represented without escaped characters.

[v0.130.0] - 2025-03-31 - Visual Workflows, Dynamic APIs, and UX Polish

Added

✨ Agent Workflow Visualization: Introduced a comprehensive system for visualizing agent workflows, including detailed step information, input/output events, and event flow. This is exposed via the API and rendered in the web UI.
🦾 Dynamic Service Discovery: A new /suite API endpoint and associated DTOs allow frontends to dynamically discover and display all available services (e.g., Agents, Threads, Tokens) with their localized names, descriptions, and icons.
🖼️ Agent Icon Support: Agents can now define a specific icon in their configuration, which is displayed in the UI and discovery endpoints.
📄 New Admin Agent Page: A dedicated administration page for agents has been added to the web UI, allowing users to browse and inspect discovered agents and their workflows.
⚙️ Refined Controller Metadata: All API controllers now include localized names, descriptions, and icons, enhancing discoverability and documentation in the OpenAPI specification.
🖥️ Workflow Visualization Components: New UI components (EventEdge, StartNode, StepNode, StopNode, Visualization) have been added to render interactive agent workflow graphs in the web application.
📝 UserMessageEvent Form Component: A dedicated component for rendering UserMessageEvent forms in the web UI has been introduced.
👤 User Settings Component: A new UserSettings component in the web UI provides quick access to user-specific options, including logout.
⚡️ Health Service Page: A basic health service page has been added to the web UI.

Changed

🔄 Unified Runner Architecture: The ApiRunner and BotRunner now inherit from a new abstract Runner base class, centralizing core application setup and lifecycle management for better consistency.
🌐 Enhanced Locale Handling in API: Agent and Thread services now consistently pass LocaleHandler to DTOs, ensuring localized data for agent names, descriptions, and other metadata is correctly returned to clients.
🗂️ Improved Module Organization: Renamed the EventPersister module for better code structure and clarity.
🧩 Updated Agent DTOs: The AgentDTO now utilizes a new AgentConfigDTO and incorporates the network_graph for a more comprehensive representation of agent capabilities.
🧑‍💻 User Data Structure: The DTO for logged-in users has been renamed from UserDTO to MyUserDTO for clearer distinction and specificity.
🎨 UI Theme Colors: Adjusted primary colors in the AI Hub theme for a refreshed visual appearance.
📊 UI Agent List: The agent list in the web UI now dynamically adapts its display (table or data view) based on available screen width, and explicitly shows agent icons and conversational status.
🚀 OIDC Authentication Flow: Improved OIDC silent renewal and error handling in the web application for a more robust and seamless authentication experience.
📦 Dependency Updates: Updated numerous dependencies across aihub_agent, aihub_api, aihub_bot, and aihub_web for improved performance, security, and stability.

Fixed

🐛 Robust Authentication Handling: Improved the TokenAndOauth2Handler to be more resilient to non-401 HTTP exceptions from individual authentication strategies, ensuring more graceful fallback and user experience.

Refactor

🧹 Codebase Simplification: Removed outdated or redundant event and agent type definitions in the web UI, streamlining the codebase and relying more heavily on OpenAPI-generated types for consistency.
🧹 Web UI Component Capitalization: Standardized component naming by changing thread/Chat.vue to Thread/Chat.vue and thread/ThreadList.vue to Thread/ThreadList.vue for consistency.
⚙️ Decorator Metadata Extraction: Refactored internal step and precondition decorators to accurately extract output events and step icons from agent methods.

Removed

🗑️ Deprecated Workflow Visualizer: Removed the old NetworkXVisualizer class, which has been entirely superseded by the new, more capable WorkflowVisualizer.

[v0.129.0] - 2025-03-28 - Introducing Bounded Loop Workflows and Enhanced Agent Examples

Added

✨ Introduced BoundedLoopAgent: A new example agent demonstrating a controlled, iterative workflow that loops a configurable number of times, showcasing advanced agent design patterns.
📄 Added BoundedLoopAgentConfig: Provides a dedicated configuration class for the BoundedLoopAgent, enabling easy customization of the loop's maximum iterations.
⚙️ Defined Custom Events for Loop Control: Introduced BeginEvent, ProcessEvent, and DecisionEvent to meticulously manage the state and flow within the BoundedLoopAgent's iterative process.
🧪 Implemented BDD Tests for Bounded Loop Workflow: Comprehensive behavior-driven development tests were added to ensure the precise functioning and iterative behavior of the BoundedLoopAgent.
🚀 Provided a trigger.py example: A simple script to easily run and observe the BoundedLoopAgent in action, aiding in understanding and demonstration.

[v0.128.0] - 2025-03-27 - Improved Embedding Reliability

Fixed

🐛 Enhanced Embedding Robustness: Implemented a recursive batch splitting strategy within the embedding process to gracefully handle validation errors that might occur during batch embedding generation, improving the reliability of vector generation for large inputs.

[v0.127.0] - 2025-03-25 - Enhanced Bot Chat Interactions

Changed

⚡️ Enhanced Typing Indicators: Implemented continuous and stoppable typing indicators for agent and OpenAI chat bots, providing a more responsive and fluid user experience during bot message processing.

[v0.126.0] - 2025-03-24 - Enhanced LLM Call Reliability

Added

✨ Configurable LLM Request Timeout: Introduced a new timeout parameter in ChatLLMConfig, enabling users to specify a maximum duration for chat-based Large Language Model requests, with a default of 30 seconds.
⚙️ Applied Timeout to Chat LLM Implementations: The newly added timeout configuration is now actively utilized by both Azure OpenAI and Self-Hosted LLM integrations, providing greater control and improved reliability for API call durations.

[v0.125.0] - 2025-03-21 - Introducing WebUI Agent and API Modernization

Added

✨ New WebUI Agent: Introduced a new WebuiAgent enabling seamless integration with OpenAI-compatible Web UI APIs, supporting streaming chat responses and configurable features like web search.
⚡️ WebUI Agent Utilities: Added dedicated utility functions and configuration for the WebuiAgent to handle Server-Sent Events (SSE) streaming and flexible setup.
📚 WebUI Agent Playground: Included a new playground example to demonstrate the WebuiAgent's functionality and ease of use.
📦 OpenAI Library Dependency: Incorporated the openai library into aihub_lib to support new integrations and future features.

Changed

🔄 OpenAI API chat_id Handling: Modified ChatCompletionRequest to move the chat_id from a top-level field into the metadata dictionary, aligning with broader API conventions.
🚫 Recursive Agent Exclusion: Enhanced the /models endpoint in the OpenAI API to allow excluding WebuiAgent types, preventing potential recursive loops during agent discovery.
🖼️ Default Image Model Update: Updated the default image generation model to DALL-E 3 for improved quality and capabilities.
🚀 Azure OpenAI API Version: Upgraded the Azure OpenAI API version in development environments to 2025-01-01-preview for access to the latest features.
📊 Performance Report Enhancement: Improved performance test reporting by adding an explicit 'Error' column, providing clearer insights into test failures.

Refactor

🧹 Function Argument Cleaning: Refactored internal handling of function arguments to filter out None values, improving robustness and consistency.
⚙️ Internal Model Naming: Streamlined the derivation of model_name within the chat service for better clarity and maintainability.
📄 Type Hint Refinement: Improved type annotations for asturn_llm_for_cost_reporting to enhance code clarity and correctness.

Changed

🚀 Streamlined Agent Event Processing: Relaxed the validation for external agent events, allowing for more flexible event distribution to agents without requiring explicit listing within a thread's agent list, enhancing dynamic agent interactions.

Refactor

🧹 Unified Bot Database Structure: Standardized MongoDB collection names to bot_conversations and bot_paths and unified the database name to aihub for the bot service. This improves database organization and consistency across services.

[v0.123.0] - 2025-03-21 - Enhanced Document Processing and Metadata Management

Added

✨ Introduced MetadataExtractor Interface: Added a new abstract class (MetadataExtractor) that allows for custom logic to extract and enrich metadata from document splits during processing, enabling more flexible and powerful document understanding.
🏷️ Extended Node Metadata: New metadata fields, REFERENCE_NAME and REFERENCE_URL, have been added to document nodes, facilitating richer attribution and precise source linking for retrieved content.

Changed

📄 Improved Markdown Parser Extensibility: The MarkdownStructuralNodeParser now integrates the new MetadataExtractor, allowing for a pluggable approach to extracting detailed metadata directly during markdown content parsing.
📊 Updated Metadata Table Display: The nodes metadata table now includes the newly introduced REFERENCE_NAME and REFERENCE_URL fields, providing comprehensive insights into document sources and references.

Refactor

🧹 Module Reorganization for Split Class: The Split dataclass, a core component in document parsing, has been moved to its own dedicated file for improved modularity and code clarity.
⚙️ Pydantic v2 Migration: The MarkdownStructuralNodeParser has been updated to align with Pydantic v2, enhancing type handling and configuration management for improved stability and future compatibility.

[v0.122.0] - 2025-03-21 - Enhanced Error Handling and System Stability

Added

✨ Introduced Robust Exception Handling: The system now proactively captures and processes ExceptionEvents originating from core services, enhancing error propagation and system stability.
🚀 Improved API Error Reporting: The Agent API (aihub_api) now correctly translates internal ExceptionEvents received from agent operations into HTTP 500 errors, providing clearer feedback on operational failures.
🛡️ Enhanced Chat Service Reliability: The Chat Service (aihub_lib) now gracefully terminates ongoing chat operations when an ExceptionEvent is encountered, preventing stalled processes and ensuring consistent state.

Changed

🔄 Extended Event Handling Signature: The stop_event mechanism within the Chat Service (aihub_lib) has been updated to explicitly recognize and store ExceptionEvents, allowing for more comprehensive error state management.

[v0.121.0] - 2025-03-20 - Enhanced Event Causality and Dispatching Reliability

Changed

🔄 Improved Event Ordering and Filtering: The system now utilizes JetStream's native sequence numbers instead of event creation timestamps for filtering and retrieving events. This change significantly enhances the precision and reliability of event causality, ensuring steps are triggered based on the exact sequence of events in the workflow.

Refactor

⚙️ Core Event Store Logic Update: The internal logic within the Dispatcher, JetStreamEventStore, and RunEventStore has been updated to fully leverage JetStream sequence numbers for all event-related operations. This provides a more robust and accurate foundation for event processing.
📚 Event Metadata Enrichment: All base events now automatically store their corresponding JetStream sequence number upon reception, providing a foundational improvement for strictly ordered event processing and dependency resolution within the agent framework.

[v0.120.0] - 2025-03-20 - Introducing Language Detection for Enhanced Processing

Added

✨ Introduced Language Detection: A new utility has been added to automatically detect the language (German, English, French, Italian, or other) of user queries using an LLM, enabling more language-aware processing capabilities.

Refactor

🧹 Streamlined IDE Module Configuration: Optimized internal .idea project files to improve module referencing and overall development environment setup.

[v0.119.0] - 2025-03-19 - API Routing Improvements

Changed

🔄 Enhanced dynamic route naming for agent event controllers to incorporate a route postfix, which improves the uniqueness and clarity of dynamically generated API endpoints.

[v0.118.0] - 2025-03-19 - Improved Event Handling and Core Updates

Added

✨ Enhanced Event Serialization: Introduced a new model_dump_json method within BaseEvent to provide robust serialization of event data to JSON strings, ensuring all data, including previously unknown fields, is fully preserved.

Changed

🧹 Refined Event Method Overrides: Added the @override decorator to the model_dump method in BaseEvent for improved code clarity and maintainability.

[v0.117.0] - 2025-03-19 - Empowering Custom Workflow Control

Added

✨ Custom Start and Stop Event Support: Introduced the ability to define and utilize custom StartEvent and StopEvent types, enabling more tailored and application-specific workflow initiation and termination for agents.
🚀 New Playground Example for Custom Events: Added a comprehensive playground example (CustomStartStopEventAgent) that demonstrates how to implement and integrate custom start and stop events into agent workflows, providing a practical guide for developers.
🧪 Dedicated Tests for Custom Workflows: Included new BDD tests to validate the functionality of custom start and stop events, ensuring reliability and providing a robust example for extending agent capabilities.

[v0.116.0] - 2025-03-19 - Enhanced Chat Control and Event Visibility

Added

✨ Enhanced Chat Interaction Control: Introduced a new subscribe_to_thread parameter in chat services, allowing users to choose between subscribing to events only from specific agents or all events within a thread for comprehensive visibility.

Changed

🔄 Broadened Default Chat Visibility: Streaming and JSON chat interactions now default to subscribing to all events within a thread, providing a more complete view of multi-agent conversations without explicit configuration.
🚀 Contextualized JSON Event Interactions: The start_json_event_interaction API now explicitly receives agent_class and agent_id, allowing for more contextual and precise handling of JSON-based conversations.
⚡️ Refined Interaction Termination Logic: Chat interactions now precisely terminate only when the primary agent sends a stop event, preventing premature endings in complex multi-agent scenarios where multiple agents might be active.
🏷️ Dynamic Model Naming for JSON Interactions: The reported model name for JSON interactions now dynamically reflects the specific agent_class/agent_id, providing clearer identification in logs and metrics.
📄 Improved Human-in-the-Loop Event Handling: Enhanced the deserialization of Human-in-the-Loop request events for increased robustness and accuracy in handling user interventions.

Fixed

🐛 Corrected display_id for Human-in-the-Loop: Ensured that display_id for Human-in-the-Loop (HITL) response events correctly reuses the display_id from the original HITL request, maintaining conversation context.

Refactor

🧹 Streamlined NATS Event Imports: Cleaned up and reorganized NATS event-related imports in ExternalEventDistributor and OpenaiService for improved code clarity and maintainability.

[v0.115.0] - 2025-03-19 - Enhanced Event Distribution

Refactor

🔄 Unified External Event Handling: The core mechanism for processing incoming user events has been significantly refactored. The WebSocketReceiver component and WSUserEvent have been renamed to ExternalEventDistributor and ExternalEvent, respectively. This change broadens the system's capability to handle user events originating from diverse external sources, including APIs, bot frameworks, and WebSockets, moving beyond a WebSocket-specific focus.
🧹 Reorganized Event Distribution Modules: Related modules and dependencies for external event distribution have been moved from the legacy sockets package into a new, more logically structured nats/distributor pathway. This improves the architectural clarity and modularity of the event handling system.
🦾 Updated API and Bot Integrations: All API endpoints and bot services responsible for receiving and processing user input have been updated to seamlessly integrate with the new ExternalEventDistributor, ensuring consistent and channel-agnostic event flow across the platform.

[v0.114.0] - 2025-03-19 - Enhanced Event Robustness and API Output Control

Added

🛡️ Robust Event Deserialization: Introduced a new mechanism for deserializing NATS events that allows a receiving application to process events even if their exact class is not locally registered. This enhances forward compatibility and inter-service communication by intelligently falling back to the most specific known parent class and preserving unknown data.
⚙️ Event Utility Functions: Added get_inheritance_depth and get_parent_classes_until_base to precisely determine event hierarchies, critical for the new robust deserialization logic.
📤 Dynamic API Output Model Generation: Implemented create_output_model to dynamically generate Pydantic models for API responses, ensuring cleaner and more controlled exposure of event data by filtering internal metadata.

Changed

🔄 Standardized Event Type Checking: Refactored all event type checks across the codebase from isinstance(event, EventType) to explicit event.is_event_type properties (e.g., event.is_start_event, event.is_control_event) on BaseEvent. This improves code readability and leverages the new robust deserialization.
📄 API Event Response Refinement: API endpoints now utilize the new output model generation to present a more focused and cleaner representation of event data in responses.
🛠️ Pydantic Model Configuration: Updated BaseEvent Pydantic configuration to allow extra fields, supporting the new robust deserialization of unknown event types without validation errors.

Refactor

🧹 Internal Event Type Handling: Consolidated and improved the internal handling of event types across dispatchers, tracing coordinators, and publishers to align with the new property-based type checking.

[v0.113.0] - 2025-03-17 - Enhanced Event Persistence and Querying

Added

✨ Event Inheritance Tracking: Introduced a new _parent_class_names computed field in BaseEvent to automatically capture the inheritance hierarchy of each event, providing deeper introspection capabilities.
📄 Persisted Event Parentage: Added an event_parents field to PersistedEventEntity to store the inheritance chain of events, enabling more flexible and robust querying based on event types.

Changed

🚀 Robust Event Querying: Updated event retrieval methods in PersistedEventEntity to leverage the new event_parents field, allowing queries to filter events based on their entire class hierarchy, leading to more resilient and accurate results.
⚙️ Simplified Local API Development: The aihub_api playground environment now defaults to using NoAuthHandler, simplifying local development and testing by removing authentication requirements.

Refactor

🧹 Streamlined Event Persistence: The core logic for persisting events has been moved from EventPersister to a new class method PersistedEventEntity.persist_event, centralizing persistence responsibilities within the entity itself.

[v0.112.0] - 2025-03-17 - Deeper Conversations: Human-in-the-Loop and Extended API Capabilities

Added

✨ Introduced Human-in-the-Loop (HITL) Capabilities: Seamlessly integrate human intervention into AI conversations, allowing for pauses to gather user input and resume interactions with the provided response.
💬 Expanded OpenAI Chat Completion Parameters: Added support for a comprehensive set of OpenAI Chat Completion request parameters, including chat_id, user, audio, frequency_penalty, function_call, functions, logit_bias, logprobs, max_completion_tokens, max_tokens, metadata, modalities, n, parallel_tool_calls, prediction, presence_penalty, reasoning_effort, response_format, seed, service_tier, stop, store, stream_options, temperature, tool_choice, tools, top_logprobs, and top_p.
🖼️ Extended OpenAI Image Generation Parameters: Enabled additional options for image generation requests, including n, quality, response_format, size, and style.
🏷️ Explicit Thread ID Assignment: Added the ability to assign explicit thread IDs when creating new conversation threads, allowing for more precise control over conversation history and state.
📈 New Human-in-the-Loop Event Retrieval: Introduced dedicated methods to retrieve specific Human-in-the-Loop request and response events from persisted history.

Changed

🔄 Refined Chat Conversation Management: Updated the core chat service logic to better handle conversation threads, supporting the continuation of existing chats and more robust linking of bot interactions to specific threads.
👤 Improved User Context in Chat Completions: Automatically populates the user field in OpenAI chat completion requests, enhancing user traceability and context within conversations.
📊 Enhanced Message History Reconstruction: Modified the process for building message history to accurately include and display Human-in-the-Loop interactions, providing a more complete conversation log.
⚡️ Streamlined OpenAI API Request Arguments: Optimized the handling of chat completion arguments by removing redundant chat_id from core function arguments before passing to the OpenAI client.
🗣️ Unified Content Building for Responses: Updated response content building to integrate potential Human-in-the-Loop questions into the final output when a stop event is triggered by an HITL request.

Fixed

⏱️ Corrected Chronological Ordering of Display Events: Ensured all persisted display events, including chat messages and Human-in-the-Loop requests, are retrieved and ordered chronologically for accurate conversation playback.

[v0.111.0] - 2025-03-17 - Optimized NATS Message Routing for Display Events

Changed

⚡️ WebSocketReceiver now leverages both NATS core and JetStream publishers, routing display events through NATS core for enhanced real-time performance and reduced latency.
🔄 Updated API and Bot services to align with the new WebSocketReceiver architecture, passing the NATS client instance during initialization.

[v0.110.0] - 2025-03-14 - Improved Bot Message Handling with Dedicated Content Extraction

Added

🚀 Unified Message Content Handling: Introduced a new ContentExtractor module that centralizes and standardizes the processing of diverse incoming message content, including text, images, and various file types from platforms like Slack and Microsoft Teams.
🖼️ Enhanced File and Image Attachment Support: The bot can now more reliably process and normalize attached files and images, converting images to base64 data URLs and handling text files to ensure seamless integration into conversations.
📄 Standardized File Information Model: Implemented a FileInfo model to provide a consistent representation of file details across different communication channels, improving the robustness of content ingestion.

Refactor

🧹 Refactored Bot Content Processing: Extracted all message content extraction logic from the main Service module into a dedicated ContentExtractor class, significantly improving code modularity, readability, and maintainability within the bot service.

[v0.109.0] - 2025-03-14 - Enhanced Chat Bot Interactions and Attachment Processing

Fixed

🐛 Resolved MS Teams Conversation Context Issues: Implemented logic to automatically clear previous conversation context for bots in Microsoft Teams when the bot is added to a new or existing chat, ensuring a fresh start and preventing stale interactions.

Changed

🖼️ Improved Image Attachment Handling: Switched to fetching and base64 encoding image attachments from Slack and Microsoft Teams. This enhancement ensures images are consistently processed as data URLs, improving compatibility and reliability for multimodal model interactions.

Refactor

🧹 Streamlined Attachment Processing: Introduced a new internal utility method _fetch_and_encode_to_base64 to centralize and simplify the fetching and encoding of file attachments, leading to cleaner and more maintainable code for handling various file types.

[v0.108.0] - 2025-03-13 - Asynchronous Preconditions for Flexible Workflow Control

Changed

⚡️ Enhanced Workflow Preconditions: Step precondition functions (precondition argument in @step decorator) can now be asynchronous (async def). This allows for more dynamic and non-blocking checks before a workflow step executes, such as waiting for external events or performing asynchronous data lookups.
🔄 Updated Dispatcher Logic: The internal dispatcher has been updated to correctly handle and await asynchronous precondition functions, ensuring seamless integration and execution of the new capability.

[v0.107.0] - 2025-03-13 - API Enhancements and Internal Utilities

Added

✨ Introduced clear_cache method for AgentService: A new static method has been added to allow clearing in-memory caches used for agent discovery, primarily beneficial for testing purposes to ensure fresh discovery requests.

Refactor

🔄 Promoted LLMStopEvent to Core Library: The LLMStopEvent has been moved from the playground experimental area into the central aihub_lib.nats.events module, making it a fully integrated and accessible part of the core event system.
🧹 Formalized GuardRejectionEvent Exposure: The GuardRejectionEvent is now directly exposed and importable from aihub_lib.nats.events, improving consistency and ease of use for guard-related event handling.
⚡️ Streamlined Event Imports: Optimized internal import paths within aihub_lib.nats.events to centralize and simplify how various event types are accessed throughout the codebase.

Removed

🗑️ Deprecated Playground Event Files: The LLMStopEvent file and its associated __init__.py have been removed from the playground directory, reflecting its migration to the core aihub_lib.

Changed

🔄 Component Version Synchronization: All core components, including aihub_agent, aihub_api, aihub_bot, aihub_lib, and aihub_pipeline, have been updated and synchronized to version v0.105.0.
📄 Documentation Formatting: Minor cosmetic adjustments were made to the Mermaid flowchart syntax within the introduction documentation for improved clarity.

[v0.104.0] - 2025-03-10 - Enhanced Bot Attachment Handling and Slack Integration

Added

✨ Introduced Slack File Attachment Support: Bots can now process image and text file attachments from Slack messages by utilizing a configured Slack OAuth token, significantly enhancing conversational capabilities with files.
🚀 Enabled Microsoft Teams File Attachment Handling: The bot now intelligently processes file attachments (including images and text files) from Microsoft Teams conversations, ensuring richer interactions.
🔑 Added Slack Token Configuration: A new slack_token field has been introduced to the bot's path-specific configuration, allowing for persistent and secure storage of Slack OAuth tokens.
⚙️ New CLI Option for Slack Token Setup: The setup_azure_bot.py script now includes a --slack-token option, simplifying the configuration of Slack OAuth tokens during bot deployment.

Changed

🔄 Refactored Conversation Message Handling: The path parameter is now consistently passed to methods responsible for adding user and bot messages to conversations, enabling more granular and path-specific message processing and credential retrieval.
🧹 Improved Attachment Processing Robustness: The bot's attachment handling has been enhanced to provide clearer logging and error handling for various attachment types, including explicitly ignoring redundant HTML attachments from Teams and warning about unsupported file types.

[v0.103.0] - 2025-03-10 - Streamlined Agent Interactions: API Endpoints and Event System Enhancements

Added

🚀 New Agent Interaction API: Introduced a new API endpoint (POST /agent/{agent_class}/{agent_id}/send_event) that allows external systems to directly initiate conversations with specific agents by sending a StartEvent and receiving a corresponding StopEvent.
⚙️ Dynamic Input Model Generation: Added a utility (create_input_model) to dynamically create Pydantic input models for API endpoints, automatically excluding generated event fields (like IDs and timestamps) for cleaner API usage.

Changed

🔄 Refined LLM Stop Event Handling: The LLMWrappingAgent now utilizes a more specific LLMStopEvent to signal the completion of LLM interactions, providing richer context than the generic StopEvent.
💬 Expanded WebSocket Event Support: The WebSocket receiver now supports StartEvent payloads within WSUserEvent, enabling more diverse interaction initiation patterns from the client side for agent interactions.
⚡️ Improved Event Signal Management: Internal ChatService logic was updated to clearly distinguish between an asynchronous stop signal (stop_signal) and the final received stop event object (stop_event), enhancing clarity and control over event streams.

Refactor

🧹 Event Stop Signal Renaming: Renamed internal _stop_event variables to _stop_signal across agent runners and performance tests for improved clarity and consistency in signaling asynchronous task completion.
📦 Centralized LLMStopEvent: Moved the LLMStopEvent definition from agent-specific folders to the aihub_lib package under a new semantic/llm namespace, centralizing event definitions and improving modularity.

[v0.102.0] - 2025-03-07 - Enhanced Document Metadata Handling

Fixed

🐛 RefDocDocument: Improved robustness of metadata property access (e.g., namespace, hash, uri, updated) by transitioning from direct dictionary access to using .get() methods with default fallback values, preventing potential KeyError exceptions for missing metadata.

Changed

⚡️ RefDocDocument: Enhanced the metadata enrichment process from DataLakeFile objects to intelligently prioritize existing metadata fields and gracefully fall back to default attributes when populating document properties like namespace, hash, updated, source, and document title.

[v0.101.0] - 2025-03-07 - Next-Gen Event Persistence and Workflow Control

Added

✨ Introduced JetStream Event Store: Replaced the Redis-based event storage with a robust NATS JetStream implementation, enabling durable, consistent, and replayable control event persistence for workflows. This significantly improves data reliability and recovery.
🚦 New Workflow Precondition Decorator (@precondition): Allows agents to define complex, data-driven preconditions for workflow steps, ensuring steps only execute when all necessary criteria are met. This enables more sophisticated and reliable workflow orchestration.
🚀 Comprehensive Performance Testing Framework: Added a dedicated playground and tools for benchmarking agent event processing, including raw NATS JetStream performance and full system throughput, providing crucial insights for scalability and optimization.
📁 Dedicated NATS Server Configuration: Included a nats.conf file for more controlled and robust NATS server deployments within Docker environments.
📊 New Redis List and Counter Utilities: Added helper methods for appending to Redis lists and atomically incrementing counters, improving data management within agent state.

Changed

🔄 Refined NATS Eventing Strategy: Separated event publishing and subscribing into distinct NATS Core (for display/ephemeral events) and JetStream (for durable control events) channels, optimizing performance and reliability based on event criticality.
💾 Enhanced JetStream Stream Durability: Configured JetStream streams to use file storage instead of memory, significantly increasing message retention and persistence.
⚙️ Optimized JetStream Stream Parameters: Increased maximum messages capacity and reduced the duplicate window for JetStream streams, improving overall stream management.
🧹 Improved Agent Step Discovery Performance: Applied caching to agent methods responsible for discovering workflow steps and events, reducing overhead and speeding up agent initialization.
⚡️ Optimized Step Execution Counting: Transitioned step execution counting in the StepStore to use Redis atomic increment operations, boosting efficiency and accuracy.
🔑 More Robust Event Parameter Keying: Implemented MD5 hashing for event parameter keys in the StepStore, ensuring more unique and reliable identification of step calls with specific event sets.
🔗 Updated Docker Compose NATS Configuration: Modernized NATS setup in Docker Compose files, utilizing a dedicated configuration file for better control and health checks.

Removed

🗑️ Deprecated Redis Event Store: The DistributedEventStore (Redis-based) has been removed, replaced entirely by the new JetStream-based event persistence.

Fixed

🐛 Graceful Shutdown for Multiprocess Agents: Improved signal handling (SIGTERM, SIGINT) in MultiprocessAgentRunner to ensure a more robust and graceful shutdown of agent processes.
🐞 Robust NATS Unsubscription: Enhanced NCSubscriber to gracefully handle ConnectionDrainingError during unsubscription, preventing potential errors during NATS client shutdown.
🛠️ Consistent Resource Cleanup: Ensured all dispatcher and subscriber resources are properly stopped and cleaned up during agent shutdown for better stability.

Added

✨ System Message Configuration for Azure Bots: Introduced the capability to define and store a default system_message during Azure bot setup, enabling better initial bot personality or instructional guidance. This message is now saved to the connected MongoDB or Cosmos DB instance.

Refactor

🧹 Relocated Azure Bot Setup Script: The setup_azure_bot.py script has been moved into its own dedicated subdirectory (aihub_bot/aihub_bot/), improving the overall organization and clarity of the bot module.

[v0.99.0] - 2025-03-07 - Enhanced Chat Context and Resilience

Added

✨ Enabled OpenAI Message Naming: Implemented support for passing a name field with chat messages to OpenAI, allowing the model to distinguish between different participants (e.g., specific users or tools) in a conversation. Names are automatically cleaned to meet API requirements.

Changed

📝 Standardized Message Naming: Messages now consistently include the sender's name property (user, bot, system) in the conversation history to improve context for advanced AI features.
📄 Enriched System Messages: System messages can now dynamically include the assistant's name using a new {assistant_name} placeholder, enabling more personalized interactions.
🤖 Agent Message Refinements: Ensured agent chat messages properly include the sender's name for better context in multi-agent or tool-using scenarios, aligning with the new message structure.

Fixed

🐛 Improved OpenAI Error Handling: Gracefully handles BadRequestError responses from OpenAI API calls, sending the detailed error message back to the user instead of failing silently.

[v0.98.0] - 2025-03-06 - Next-Gen Chat: Multimodal Support and Enhanced Slack Integration

Added

✨ Multimodal Message Support: Introduced the ability for the bot to process and respond to messages containing various content types, including plain text, image URLs, and content from attached text files, enabling richer and more versatile conversations.
💬 Content Data Model: A new Content embedded document was added to the conversation message structure, allowing messages to store multiple content blocks with distinct types (e.g., text, image_url).
🚀 Slack Direct Message Detection: Implemented new logic to accurately distinguish between direct messages and channel messages in Slack, improving the bot's contextual awareness for responses.
📌 Conversation Mention Tracking: Introduced an is_mentioned flag within conversation entities to track whether the bot has been explicitly mentioned in a Slack channel, enabling more intelligent and targeted responses.

Changed

🔄 Standardized Slack Message Handling: Refactored the core Slack message processing logic across all bot types into a unified service method (handle_slack_message), which now orchestrates the handling of direct messages, channel mentions, and continued conversations.
📄 Core Message Structure: Updated the internal Message entity to support a list of Content objects, replacing the previous single string content field, which is foundational for multimodal input capabilities.
🗣️ LLM Integration for Multimodal Input: Modified the integration layers for both Agent and OpenAI bots to correctly parse and pass multimodal content (text and images) from user messages to the underlying Large Language Model services.

Refactor

🧹 Centralized Slack Logic: Consolidated various Slack-specific checks and message preparation steps into a single, comprehensive handle_slack_message function, enhancing code clarity and maintainability.
⚙️ Internal Helper Renames: Renamed several internal message conversion helper methods to follow private naming conventions (e.g., message_to_chat_message to _message_to_chat_message), improving encapsulation.

[v0.97.0] - 2025-03-06 - LLM Configuration Enhancements

Added

✨ Enhanced LLM System Prompt Control: The cost_reporting_llm context manager in ChatLLMConfig now supports an optional system_prompt parameter, allowing for direct configuration of system messages for Large Language Models within cost-tracked sessions.

[v0.96.0] - 2025-03-06 - Enhanced Bot Message Handling

Fixed

🐛 Improved Bot Message Sending: The bot service now gracefully handles empty message buffers, preventing unnecessary activity updates and enhancing the reliability of message delivery.

[v0.95.0] - 2025-03-06 - Improved Chatbot Interactions and Streamlined Environment Setup

Added

📄 Standardized Testing Environment Configurations: Introduced new .env files for testing environments, ensuring consistent setup alongside development.

Changed

💬 Enhanced Chatbot User Feedback: Implemented "typing" indicators for all chatbot types (Agent, Stream Agent, OpenAI, and Stream OpenAI), providing real-time feedback during response generation.
⚙️ Refined Development Environment Variables: Updated the structure of development .env files to directly use AZURE_SUBSCRIPTION_ID for more explicit Azure resource identification.

Refactor

🧹 Optimized Chat Service Message Sending: Inlined the text sending logic within the send_response_stream method, improving the internal organization and maintainability of the chat service.

[v0.94.0] - 2025-03-05 - Enhanced Bot Streaming and Configuration Flexibility

Changed

✨ Improved Bot Streaming Response Handling: Enhanced the bot's ability to send long streaming responses by introducing a more robust chunking and update mechanism, including graceful handling for messages that exceed length limits to ensure continuous user experience.
🔧 Flexible Azure App Naming: The APP_NAME configuration for Azure deployments now supports hyphens, providing greater flexibility for naming conventions.

Fixed

🐛 Agent Stream Timeout Robustness: Resolved an issue where agent streaming responses could lead to indefinite waits by implementing a timeout mechanism with graceful handling for stream termination, improving overall stability.

[v0.93.0] - 2025-03-05 - Core Component Version Alignment

Changed

🔄 Component Version Synchronization: Updated the project version and internal aihub_lib dependency tags across all aihub_agent, aihub_api, aihub_bot, aihub_lib, and aihub_pipeline components to v0.93.0 to ensure full system compatibility and release alignment.

[v0.92.0] - 2025-03-05 - Intelligent Context Handling for RAG Agents

Added

🦾 Introduced Context Sufficiency Guard: A new context_sufficient_guard has been implemented to assess whether the retrieved context is adequate for answering a user's query within RAG agents. This enhances response quality by identifying and managing cases where context is insufficient.
⚙️ Configurable Context Sufficiency Check: Added a new check_context_sufficiency option to the RAGAgentConfig, allowing users to enable or disable the context sufficiency guard based on their requirements.
💬 New Context State Events: Introduced ContextSufficientEvent and ContextInsufficientEvent to clearly signal the outcome of the context sufficiency check within the agent's workflow.
🌐 Internationalized Context Guard Prompts: Added multi-language support (German, English, French, Italian) for the new context sufficiency guard's prompts, improving global accessibility.
🚀 Agent-in-the-Loop Event Categories: New event categories have been introduced under aihub_lib/aihub_lib/nats/events/agent_in_the_loop for managing exceptions, requests, and responses related to agent-in-the-loop functionality, laying groundwork for advanced human-in-the-loop workflows.

Changed

🔄 Improved RAG Agent Fallback Behavior: The RAGAgent now gracefully handles scenarios where the retrieved context is deemed insufficient by providing a fallback response, ensuring a more robust user experience.
✨ Streamlined Few-Shot Guard Acceptance: Simplified the FewShotAcceptEvent creation in the few_shot_guard_step, removing redundant success=True parameter as it is implied by the event type.

[v0.91.0] - 2025-03-05 - Parallel Agents, GPU Support, and Enhanced Observability

Added

🚀 Multiprocess Agent Runner: Introduced a new MultiprocessAgentRunner to enable horizontal scaling and true parallelism for agents by running multiple instances across separate processes.
⚡️ GPU-Accelerated Local Development: Added docker-compose-gpu.yml to facilitate local development and testing with GPU-enabled infrastructure, including local LLM and embedding models.
⚙️ Developer Convenience Configurations: Included new PyCharm run configurations to simplify the setup and execution of tests and core components (API, Agent, Bots, Lib) for an improved developer experience.

Changed

🔄 Event Store Caching Improvement: Refined the DistributedEventStore to directly cache and retrieve deserialized ControlEvent objects, enhancing type safety and internal consistency.
📈 Enhanced Logging Format: Updated the default logging formatter to include timestamps with milliseconds, providing more precise temporal context for debugging and monitoring.
📊 Tracing Span Management: Adjusted the tracing logic for successful agent steps, allowing for more flexible span management and potentially richer trace data capture.

Fixed

🐛 Semantic Event Serialization: Corrected an issue where optional attributes in semantic tracing events were not properly omitted if None, and ensured LLM_INVOCATION_PARAMETERS are correctly serialized to JSON.
🪄 Bot Module Import: Fixed a minor import path for OpenaiChatController in the aihub_bot playground.

Removed

🗑️ Deprecated Docker Compose File: Removed the non-functional docker-compose-open-webui.yml, cleaning up the project structure.

[v0.90.0] - 2025-03-04 - Improved Observability and Code Structure

Added

⚙️ Configurable Logging Levels: Introduced LoggingConfig in aihub_lib to allow setting log levels dynamically via environment variables, enhancing debugging and operational control.
⚡️ Automatic Playground Logging: Enabled enable_logging() by default in aihub_api and aihub_bot playground environments, providing immediate visibility into test and development flows.

Changed

🚀 Optimized Trace Export: Switched OpenTelemetry trace export from SimpleSpanProcessor to BatchSpanProcessor in aihub_agent, significantly improving trace reliability and performance by batching exports.
📈 Enhanced Logging Configuration: The enable_logging function now supports an additional lib_level argument and respects environment variable overrides for log levels, offering more granular control over library logging.

Refactor

🧹 API Import Cleanup: Removed numerous unused imports across various aihub_api controllers and DTOs, reducing clutter and improving code readability.
🧪 Test Suite Clarity: Renamed test functions in FewShotAgent and AgentInTheLoopAgent playgrounds for improved clarity and maintainability of test scenarios.
🔄 Standardized OpenAI URLs: Centralized OpenAI base URLs in aihub_api playground configuration for easier management and consistency.
📄 Simplified NATS Event Imports: Refactored the import structure for AgentInTheLoop events within aihub_lib.nats.events to streamline access and reduce nesting.

Removed

🗑️ Redundant NATS Event Modules: Deleted empty __init__.py files previously used for re-exporting AgentInTheLoop events, simplifying the module structure.

[v0.89.0] - 2025-03-03 - Enhanced Agent Dispatching with Idempotency

Added

✨ Introduced Idempotent Agent Step Execution: Agent steps now include a robust mechanism to prevent duplicate executions within the same run when called with identical input events. This significantly enhances the reliability and efficiency of agent dispatching by avoiding redundant processing.

[v0.88.0] - 2025-03-03 - Streamlined Communication and Resource Management

Changed

🚀 Improved NATS JetStream Configuration: Enhanced the NATS JetStream client setup in AgentRunner with a 60-second timeout and increased publish_async_max_pending to 10,000 messages, boosting communication robustness for event publishing.
⚡️ Enhanced NATS Publishing Reliability: Implemented a robust retry mechanism with timeouts and message deduplication (using Nats-Msg-Id headers) for all NATS JetStream publications, significantly improving the reliability of event delivery.
⚙️ Optimized NATS Stream Management: Configured NATS streams to use MEMORY storage instead of FILE for improved performance. New policies for message limits (1,000,000), discarding old messages, a 30-day message age limit, and a 1-hour duplicate window were also introduced, optimizing resource usage and data retention.
🐳 Updated NATS Docker Image: Switched the NATS service in docker-compose.yml from nats:latest to the more lightweight nats:alpine image, reducing resource consumption for local deployments.

Refactor

🧹 Standardized Pre-Commit Hook Order: Reordered the pr-ready Makefile target across all core services (aihub_agent, aihub_api, aihub_bot, aihub_lib, aihub_pipeline) to consistently run formatting before linting, streamlining development workflows.

[v0.87.0] - 2025-03-03 - Infrastructure Upgrade: Redis for Distributed State & Workflow Optimizations

Added

✨ Redis for Distributed State: Integrated Redis as the primary backend for distributed event and step storage across agents, significantly enhancing performance, scalability, and simplifying state management.
⚙️ Redis Configuration: Introduced a new RedisConfig class within aihub_lib for streamlined management of Redis connection settings.
💾 Redis Docker Service: Added a dedicated Redis service to the docker-compose.yml for easier local development and testing environment setup.
🚀 Telemetry Header Caching: Implemented a TTLCache in the Dispatcher to cache telemetry headers, optimizing performance by reducing redundant data fetches during step execution.

Changed

🛠️ CI/CD Linting Step: Incorporated a lint command into the pr-ready Makefile targets for all microservices, reinforcing code quality standards prior to pull requests.
📈 NATS Stream Durability: Switched NATS stream storage from in-memory (MEMORY) to file-based (FILE), improving data durability and persistence for event streams.
🐳 NATS Docker Image Update: Updated the NATS Docker image to nats:latest and enabled verbose/debug logging (-D, -V) for better operational insights.

Refactor

🧹 Unified State Management Logic: Reworked core components including StoreBase, DistributedEventStore, StepStore, and various Context classes (BaseContext, RunContext, ThreadContext) to unify their distributed state management under Redis, simplifying the overall architecture.
🧰 Optimized Step Readiness Check: Enhanced the Dispatcher's step readiness logic to precisely retrieve only the necessary input events for step execution, leading to more efficient data fetching.
🔄 Streamlined Agent Event Imports: Cleaned up various agent files by removing explicit StartEvent imports where not directly consumed, reflecting a more consolidated approach to initial event handling within the agent framework.

[v0.86.0] - 2025-03-03 - Streamlined Bot Services

Refactor

🧹 Refined Bot System Message Retrieval: Removed redundant calls to get_system_message from StreamAgentChatBot, OpenaiChatBot, and StreamOpenaiChatBot to centralize and simplify how system messages are handled within bot services.

Changed

🔄 Updated Core Component Versions: Synchronized the versions of aihub_agent, aihub_api, aihub_bot, aihub_lib, and aihub_pipeline to v0.85.0 for consistent deployment and development.
📄 Enhanced Document Intelligence Loader: Adapted the Document Intelligence Loader to align with the latest API specifications by updating the parameter name from analyze_request to body for improved compatibility.

Removed

🗑️ Removed LimitChatHistoryWithContextEvent: The event previously used for explicitly limiting chat messages and context within the FewShotAgent has been removed.

[v0.83.0] - 2025-02-27 - Core Performance and Stability Improvements

Changed

⚡️ Optimized NATS JetStream Storage: Switched the default storage type for all NATS JetStream Key-Value stores (including run and thread contexts) and general streams from file-based (FILE) to in-memory (MEMORY). This change significantly enhances performance and responsiveness of context and stream operations.
⚙️ Enhanced KV Store Creation Robustness: Improved the error handling for Key-Value store creation, ensuring that the system more gracefully attempts to retrieve an existing store if creation fails, leading to increased stability.

Added

✨ Implemented caching for event store lookups within the DistributedEventStore, significantly improving performance by reducing redundant data fetches for frequently accessed run-specific information.

Changed

⚡️ Refined the step execution count logic in the Dispatcher to conditionally increment based on the presence of a _max_executions_per_run attribute, allowing for more precise control over step retries and execution limits.

Removed

🗑️ Deprecated and removed the retry_operation method from StoreBase, streamlining the core store interface.
🧹 Removed mutex-based synchronization and retry configuration attributes (max_retries, base_backoff, and synchronized_update method) from BaseContext, simplifying the underlying NATS context management.

[v0.81.0] - 2025-02-27 - Enhanced Reliability and Concurrency

Changed

⚡️ Overhauled event and step execution management for enhanced concurrency and data consistency in distributed environments. This new approach leverages unique key storage in NATS JetStream, effectively eliminating race conditions previously addressed with mutexes.
🔄 Optimized event access within the Dispatcher by pre-fetching all relevant events before checking step readiness and executing steps, improving performance and ensuring consistent state snapshots.

Fixed

🐛 Improved handling for existing NATS JetStream KeyValue stores by explicitly catching "already in use" errors during store creation, preventing exceptions in high-concurrency scenarios.

Refactor

⚙️ Strengthened NATS JetStream KeyValue store resilience by increasing maximum retry attempts (from 5 to 10) and fine-tuning the backoff strategy for transient connection or operation errors.
🧹 Enhanced internal StoreBase utilities, introducing a dedicated method for storing raw byte values and generalizing the retry mechanism for various operations across the agent.

[v0.80.0] - 2025-02-27 - Enhanced Localization Capabilities

Changed

🌍 Improved Internationalization (i18n): The LocaleHandler now supports passing additional arguments for dynamic content interpolation within translated strings, enabling more flexible and context-aware localized messages.

[v0.79.0] - 2025-02-27 - Enhanced Data Consistency with Atomic Store Operations

Added

✨ Introduced Atomic and Synchronized Updates: Implemented new atomic_operation and synchronized_update methods in StoreBase and BaseContext to ensure safe, concurrent data modifications in JetStream KV stores, effectively preventing race conditions.
⚙️ Utility Methods for KV Stores: Added get_value, put_json_value, and get_json_value to StoreBase, simplifying common data retrieval and storage patterns with optional transformations and JSON handling.

Changed

🔄 Improved Event Store Concurrency: The DistributedEventStore now leverages the new synchronized update mechanisms for adding events, significantly improving reliability and data consistency under concurrent operations.
⚡️ Enhanced Step Execution Tracking: The DistributedStepStore now uses synchronized updates for increment_execution_count and simplified checks for is_run_crashed, ensuring accurate and reliable tracking of step executions in highly concurrent environments.
🧹 Refined KV Store Key Retrieval: Key-value store listing methods across StoreBase and BaseContext now automatically filter out internal mutex keys, providing cleaner and more relevant results.

[v0.78.0] - 2025-02-27 - Streamlined Azure AI Search Configuration

Changed

⚡️ Simplified Azure AI Search Vector Store Creation: The create_azure_ai_search_vector_store function now utilizes a default value for semantic_configuration_name, which streamlines the configuration process and simplifies the internal logic for setting up Azure AI Search vector stores.

[v0.77.0] - 2025-02-26 - Enhanced Message Processing Reliability

Refactor

⚡️ Improved NATS JetStream Message Acknowledgment: Refactored the JSSubscriber to acknowledge messages immediately upon receipt, preventing re-delivery of messages if subsequent processing tasks encounter issues. This enhances the reliability and predictability of message consumption.

[v0.76.0] - 2025-02-26 - NATS Message Processing Refinement

Changed

🔄 Adjusted NATS JetStream Acknowledgment Logic: The order of message acknowledgment has been updated in the JSSubscriber to occur before the message handler is executed. This change optimizes message flow and prevents potential re-delivery of messages in scenarios where the handler might encounter an error after initial processing has begun.

[v0.75.0] - 2025-02-26 - Enhanced Asynchronous NATS Message Processing

Changed

⚡️ Improved NATS JetStream Message Processing: The JSSubscriber now handles incoming messages asynchronously by creating a separate task for each message. This prevents the main message loop from being blocked by slow handlers, significantly enhancing concurrency and system resilience.
🚀 Optimized NATS Core Message Handling: The NCSubscriber has been updated to process incoming messages in a non-blocking fashion. This change ensures that message handling is more robust and efficient, improving overall system responsiveness.

[v0.74.0] - 2025-02-26 - Enhanced Document Metadata Handling

Fixed

🐛 Corrected Document Title Extraction: Ensured that the document title is accurately extracted from the DataLakeFile URI when enriching RefDocDocument metadata, preventing incorrect title assignments.

[v0.73.0] - 2025-02-26 - Core Stability Enhancements and Version Alignment

Fixed

🐛 WebSocket Message Handling: Enhanced robustness of the WebSocketReceiver to gracefully handle cases where a start event might not contain a messages attribute, preventing potential runtime errors.

Changed

🔄 Project-wide Version Bump: Aligned all core microservices and internal dependencies to version v0.73.0 for consistent build and deployment.

[v0.72.0] - 2025-02-26 - Enhanced Document Metadata for RAG

Added

✨ Enriched RefDocDocument Metadata: Improved document processing in the pipeline by automatically extracting and adding the document title and source URI to RefDocDocument instances. This enhancement provides richer, more granular context for Retrieval-Augmented Generation (RAG) applications.

[v0.71.0] - 2025-02-26 - Improved Self-Hosted LLM Compatibility

Changed

⚙️ Enhanced Self-Hosted LLM Tokenizer Compatibility: Updated the self-hosted LLM configuration to correctly infer base model names by stripping additional suffixes (-bnb and -4bit), improving compatibility with various quantized model formats for tokenization.

[v0.70.0] - 2025-02-26 - Expanded Azure OpenAI Authentication Options

Changed

🔑 Enhanced Azure OpenAI LLM Authentication: Enabled direct API key authentication for Azure OpenAI Large Language Models, providing a flexible alternative to Azure AD for resource access.

[v0.69.0] - 2025-02-25 - Enhanced Build Stability and CI Efficiency

Changed

⚡️ Optimized CI Workflow: Reordered Python and Poetry setup steps and introduced Poetry caching in the add-tag.yml GitHub Actions workflow to significantly improve build times and efficiency.
🦾 Strengthened Dependency Management in CI: Added explicit poetry lock and poetry install steps to GitHub Actions workflows, ensuring consistent and reliable dependency resolution and installation across all CI runs.
🧹 Streamlined Build Dependencies: Removed direct pip install commands for global tools like tomlkit in CI, now relying entirely on Poetry for comprehensive dependency management.

[v0.68.0] - 2025-02-25 - Enhanced Self-Hosted LLM Configuration

Changed

⚡️ Improved Self-Hosted LLM Tokenizer Loading: Enhanced the tokenizer property within SelfHostedLLMConfig to correctly load tokenizers for various self-hosted models by stripping common suffixes (like -GGUF and -AWQ) from model names and ensuring the proper encoding function is returned.

[v0.67.0] - 2025-02-25 - Build System Enhancements and Poetry Integration

Refactor

🧹 Renamed Dependency Management Script: The script for switching between local and remote cores has been renamed from switch_dependency.py to switch_dependencies.py for improved clarity.
⚡️ Poetry Integration for Development Scripts: Updated Makefile commands to utilize poetry run for executing core dependency switching scripts, streamlining the development workflow and aligning with Poetry's project management practices.

[v0.66.0] - 2025-02-25 - Enhanced Authentication, UI Overhaul, and API Clarity

Added

🖼️ User Profile Images: Implemented fetching and displaying user profile images from Azure AD, enhancing the user experience in the UI.
🔑 Flexible Authentication Handler: Introduced TokenAndOauth2Handler to enable combined use of Bearer token and OAuth2 authentication strategies, offering more versatile security configurations.
💾 Persistent User Preferences: Added UserEntity for storing user-specific data like favorite modules, laying the groundwork for personalized experiences.
⚡️ Client SDK Generation: Integrated OpenAPI-TS to automatically generate a robust TypeScript SDK for the frontend, streamlining API interactions and ensuring type safety.
🎨 New Application Layouts: Designed and implemented new default and anonymous Nuxt layouts, providing a more structured and visually consistent user interface.
👤 User Bar Component: Introduced a dedicated user bar component in the UI, displaying user information and integrating a dark mode toggle.
🚀 Redesigned Login Experience: Overhauled the login page with a modern, two-panel design and prominent Microsoft login integration.
🔄 OIDC Silent Token Renewal: Added support for silent token renewal in OpenID Connect (OIDC) to maintain seamless user sessions without visible re-authentication.
🌐 Open Web UI Integration: Enabled direct embedding of Open Web UI as an iframe within the application, providing a unified access point.
📊 User Data Store: Created a new Pinia store for managing user-related data, including profile information fetched via the API.
📄 Structured Health Response: Defined a HealthResponse DTO for the API's health endpoint, providing a clearer and standardized contract.

Changed

🔒 Standardized Security Dependencies: Migrated API routes to use FastAPI's Security instead of Depends for authentication, improving consistency and clarity in security definitions.
🛠️ Refined User Information Retrieval: Updated the API's get_user endpoint to leverage a dedicated service method for fetching comprehensive user details, including the newly added profile image.
💅 UI Framework Transition: Transitioned the web application's primary styling and component library from Shadcn UI to PrimeVue, resulting in a refreshed look and feel.
🏗️ Nuxt Layout Integration: Refactored the web application to fully utilize Nuxt Layouts, establishing a consistent structural foundation across pages.
📈 Improved OpenAPI Generation: Enhanced OpenAPI schema generation to include operation_id for API routes, facilitating more accurate and usable client SDKs.
🎨 Updated PrimeVue Theme: Applied a new zinc color palette and configured dark mode for the PrimeVue theme, aligning with updated branding and improving visual consistency.

Removed

🗑️ Deprecated Multi-Authentication Handler: Removed the generic MultiAuthHandler in favor of the more specialized and explicit TokenAndOauth2Handler.
🧹 Shadcn UI Components: Eliminated all Shadcn UI components, styling, and related configurations from the web application as part of the UI framework transition.
📦 Unused BSON Dependency: Cleaned up the aihub_agent module by removing an unused bson dependency.

Refactor

⚙️ Component Organization: Reorganized and renamed the chat component (Chat.vue to ThreadChat.vue) and moved it into a dedicated thread subdirectory for better module clarity.

[v0.65.0] - 2025-02-25 - Core Agent LLM Flexibility

Changed

🔄 Enhanced LLMWrappingAgent Configuration: The LLMWrappingAgentConfig has been updated to use the more generic ChatLLMConfig instead of the provider-specific AzureOpenAILLMConfig. This modification provides greater flexibility, enabling the LLM Wrapping Agent to integrate seamlessly with a broader range of chat LLM providers.

[v0.64.0] - 2025-02-24 - Improved Core Robustness and System Alignment

Fixed

🐛 Enhanced Locale Handling: Ensured that the system's locale context reliably defaults to a predefined value when no specific locale is provided, improving overall stability and preventing potential issues.

[v0.63.0] - 2025-02-21 - Refined Guards and Enhanced Testability

Changed

🔄 Improved Few-Shot Guard Prompting: The Few-Shot Guard has been updated with a refined internal prompt that strictly enforces JSON-formatted output for guard responses, enhancing their reliability and programmatic parsing.
⚡️ Enhanced AgentTestRunner: The get_events and get_events_of_type methods in AgentTestRunner now support an event_filter parameter, allowing for more granular observation and testing of events based on their specific event type.

[v0.62.0] - 2025-02-20 - Streamlined Development and Enhanced CI/CD

Changed

🚀 Improved CI/CD Stability: The Docker service health check in the backend test action has been reordered to occur after Python dependency installation, enhancing the reliability of automated tests.
✨ Simplified NoAuth Configuration: The NoAuthHandler now provides default values for user NAME, EMAIL, and ROLES, reducing the need for explicit configuration in local development environments.

[v0.61.0] - 2025-02-20 - Major Pipeline Configuration Overhaul

Added

✨ Explicit Data Lake Resource: Introduced the DataLakeResource to clearly define and manage data lake container and directory configurations within pipelines, enhancing configuration clarity.
➕ Dedicated Document Store Resource: Added the DocStoreResource to provide explicit configuration for document store names and associated namespaces, improving resource management.

Changed

💡 Flexible Asset Key Inputs: Asset factories for documents and nodes now support AssetKey objects as input for their respective data lake and document keys, providing greater flexibility in asset dependency definition.
⚙️ Azure AI Search Configuration: Enhanced the AzureAISearchVectorStoreFactory to correctly handle None values for semantic_configuration_name, allowing for more flexible vector store deployments without unnecessary parameters.
📄 Document Content Representation: Updated the display of RefDocDocument metadata to consistently use get_content() instead of get_text(), ensuring a more accurate representation of indexed document content in pipeline lineage.
⬆️ Improved Milvus Vector Store Documentation: Added comprehensive documentation, including detailed examples and docstrings, to the MilvusVectorStoreResource for clearer usage guidelines.

Refactor

🔄 Centralized Resource Configuration: Overhauled resource definitions by removing the generic NamespaceResource and introducing dedicated, explicit resources like DataLakeResource and DocStoreResource, enhancing clarity and maintainability of pipeline configurations.
🧹 Simplified LLM Integration: Streamlined the setup for Large Language Models and Embedding Models by removing the LlmHandlerResource and enabling direct configuration via specific configuration objects (e.g., AzureOpenAILLMConfig).
🏗️ Improved Pipeline Naming: Simplified job and asset naming conventions by removing the customer_name parameter, making asset and job names more concise and consistent.
⚙️ Refined Data Lake Operations: Updated data lake fetching and versioning operations to directly use container and directory names, aligning with the new explicit resource configuration model.
🔗 Modernized MongoDB Connection: The MongoDB connection utility now directly accepts a specific database name, improving connection clarity and robustness.

Removed

🗑️ Deprecated Generic Namespace Resource: Eliminated the overarching NamespaceResource, simplifying the resource dependency graph and enforcing more specific resource configurations.
➖ Removed LLM Handler Abstraction: The LlmHandlerResource was removed, favoring a direct, configurable approach to LLM and embedding model instantiation.
🧹 Obsolete Utility Functions: Discarded various utility functions related to customer and namespace path handling, streamlining the codebase after the resource configuration overhaul.

[v0.60.0] - 2025-02-20 - Build Action Configuration Update

Changed

🔄 Adjusted build_image GitHub Action: The file input for the build_image action no longer has a default value. This change requires users to explicitly specify the Dockerfile path, promoting clearer and more explicit build configurations.

[v0.59.0] - 2025-02-20 - Minor Infrastructure Enhancements

Changed

🛠️ Improved Backend Testing Visibility: Added docker compose ps command to the backend test action, providing better insights into service health during automated testing workflows.
🔄 Updated LLM Inference Server: Upgraded the llama.cpp Docker image to the latest server tag, ensuring access to the most recent optimizations and features from the underlying LLM inference backend.

[v0.58.0] - 2025-02-20 - Enhanced Image Building Flexibility

Added

✨ Improved build_image GitHub Action: The build_image action now supports specifying a custom Dockerfile path via a new file input, providing greater flexibility for projects with non-standard Dockerfile locations.

[v0.57.0] - 2025-02-20 - Enhanced User Authentication and Context Management

Added

✨ OpenWebUI Authentication Handler: Introduced a new authentication handler (OpenWebuiAuthHandler) for seamless integration with the OpenWebUI frontend, providing robust user identity management for deployments utilizing OpenWebUI.
🧪 Fake User Utility for Testing: Added a fake_user() utility to simplify the creation of authenticated user contexts, streamlining development and testing of agents and services.

Changed

🔄 Enhanced User Context Propagation: Updated the UserMessageEvent and core chat services (ChatService, WebSocketReceiver) to pass a rich AuthenticatedUser object instead of a simple user ID, providing more comprehensive user information throughout the system.
⚡️ API Playground Default Authentication: The aihub_api development playground now defaults to using the OpenWebuiAuthHandler, demonstrating its integration with the new authentication method.

[v0.56.0] - 2025-02-20 - Unified Core Version Update and Release Automation

Changed

🚀 Core Component Version Upgrade: All core components, including aihub_lib, aihub_agent, aihub_api, aihub_bot, and aihub_pipeline, have been updated to version v0.56.0, ensuring consistency across the ecosystem.
🔗 Dependency Alignment: Microservices now depend on the latest aihub_lib version v0.56.0, ensuring all services benefit from the most recent library improvements.
⚙️ Streamlined Release Automation: The release workflow (add-tag.yml) has been enhanced to automatically update poetry.lock files and commit Makefile changes, improving the robustness and consistency of new releases.
🛠️ Updated Default Remote Core Tag: The Makefile now defaults to v0.56.0 for the remote core tag, simplifying setup for new environments.

[v0.55.0] - 2025-02-20 - Unified AI-Hub API: OpenAI Standard for Agents and Enhanced Conversational Flows

Added

🚀 OpenAI API Compatibility for Agents: AI-Hub agents can now be discovered and interacted with using standard OpenAI API endpoints (/openai/models and /openai/chat/completions). This allows for seamless integration of AI-Hub agents into existing OpenAI-compatible applications and workflows.
💡 Conversational Agent Identification: A new is_conversational flag has been added to agent discovery responses, allowing client applications to easily identify agents designed for chat-based interactions.
✨ Dedicated User Message Event: Introduced UserMessageEvent as a specific event type for user-initiated conversational inputs. This event now encapsulates chat messages and locale settings, providing a clearer and more structured way to handle user queries.

Changed

🔄 Agent Input Event Types: Agent workflow steps that process user conversational input have been updated to exclusively accept UserMessageEvent, ensuring consistency and clarity in handling user messages.
⚡️ Enhanced User Query Tracing: The tracing system now directly utilizes the user_query property of UserMessageEvent for improved accuracy and readability of span information in conversational turns.

Removed

🗑️ Deprecated Chat Endpoints: The standalone /chat/completions API endpoints have been removed. All chat interaction functionality is now consolidated under the new OpenAI-compatible API endpoints.

Refactor

🧹 Refined Start Event: The StartEvent has been streamlined to serve as a generic workflow trigger, decoupling it from chat-specific message content and locale.
📂 Logical Chat Message Grouping: AssistantChatMessage and UserChatMessage definitions have been relocated to aihub_lib/nats/events/user/content for better logical organization.

[v0.54.0] - 2025-02-19 - Unified Authentication, API Tokens, and Chatbot Expansion

Added

🦾 New aihub_bot Module: Introduced a dedicated module for Azure Bot Service integration, enabling AI agents to be seamlessly exposed and interacted with through various chat channels like Microsoft Teams, Slack, and WebChat.
✨ API Token Management: Launched new API endpoints (/tokens) and backend services to allow users to create, list, and revoke API tokens for secure and granular access control.
🔑 Flexible Authentication Architecture: Implemented a new abstract AuthHandler and MultiAuthHandler to support multiple authentication strategies (OAuth2, API Tokens, Development/NoAuth) within the API, significantly improving extensibility and security.
🛠️ Development User Information Provider: Added DevUserInformationProvider for simplified user authentication in development and testing environments, making local setup more convenient.
🧪 Comprehensive API Test Suite: Introduced extensive new pytest-based integration and unit tests across core API functionalities (agents, chat, authentication, threads, tokens, users, health, internationalization), replacing outdated .http client test files and enhancing API reliability.
🚀 Enhanced Simulated Agent Test Runners: Improved both API and Bot test runners to better simulate agent behavior, including the handling of stop_events, for more robust and accurate testing environments.
⚙️ API-Specific Configuration: Introduced ApiConfig for managing API-specific settings independently from general Azure configurations, resulting in a cleaner and more modular setup.
🔗 ASGI Request Adapter for Testing: Added ASGIAdapter to enable seamless internal HTTP requests within test environments for FastAPI applications, which significantly improves test reliability and speed by avoiding actual network calls.
📄 Utility for API Token Generation: Provided a new Python script (generate_api_token.py) to easily generate API tokens for administrative or internal use.

Changed

🔄 Standardized Agent Event Specifications: Renamed StartEventSpecs to EventSpecs and extended the agent discovery mechanism to include stop_events, providing a more complete and accurate representation of agent lifecycle.
🔒 Agent Access Control: Enhanced the AuthenticatedUser model to allow users with the "AllAgents" role to access any agent, simplifying permission management for administrators.
🗺️ Updated Architectural Overview: Revised documentation (in aihub_doc) to explicitly reflect the integration of Chatbots as a core interface layer within the AI-Hub architecture, outlining their design and technical implementation.
🔗 Thread Creation API Route: Modified the API route for creating threads from /{thread_id} to /, aligning with RESTful design principles for improved consistency.
👤 User Information Retrieval: Updated the user information retrieval mechanism in aihub_api to leverage a multi-strategy provider, enabling fetching user details from Azure AD, API tokens, or development configurations based on the active authentication.
🛂 Health Endpoint Authentication: The health endpoint (/health) now requires authentication, leveraging the new flexible authentication handlers to ensure consistency across API endpoints.

Fixed

🐛 MongoDB Connection Closing: Ensured that the MongoDB connection is properly disconnected in the API runner's lifetime_manager, preventing potential resource leaks during application shutdown.
🐞 Thread User ID Access: Corrected a minor bug in thread management where user IDs were inconsistently accessed as user_id instead of id.

Refactor

🧹 Unified Authentication Dependency: Replaced specific use_oauth2_user and use_no_auth_user FastAPI dependencies with a generic AuthHandler interface, providing a consistent and extensible way to integrate various authentication methods across API controllers.
📦 Authentication Module Restructuring: Reorganized and renamed authentication-related modules (e.g., no_auth to NoAuthHandler, oauth2 to OAuth2AuthHandler) for improved clarity and maintainability within the authentication system.
⚙️ Core Configuration Separation: Separated general Azure base configuration (AzureBaseConfig) from API-specific settings (ApiConfig) to create a more modular and explicit configuration structure, enhancing manageability.
🗃️ Simplified Document Database Access: Streamlined database access for RefDoc entities by removing the redundant organization_shortname parameter from query methods, simplifying how reference documents are retrieved.
🧹 CI/CD Workflow Adjustments: Performed minor adjustments to GitHub Actions workflows, including Python dependency caching and poetry lock behavior, for improved reliability and performance of backend linting and testing processes.

Removed

🗑️ Deprecated Authentication Dependencies: Eliminated the outdated use_oauth2_user and use_no_auth_user functions, which have been fully superseded by the new modular authentication handler system.
🗑️ Legacy API Token Entity: Removed the old AccessToken database entity, which has been replaced by the more robust and flexible BearerToken model for API token persistence.
🗑️ Outdated HTTP Client Tests: Removed several legacy .http client test files from the aihub_api playground, as their functionality is now covered by the new, comprehensive pytest test suite.

[v0.53.0] - 2025-02-19 - Enhanced Security & Scalable Platform Foundations

Added

✨ New API Token Management: Introduced comprehensive API endpoints and a dedicated BearerToken entity for creating, listing, and revoking API tokens, enabling secure programmatic access to the platform.
🚀 Flexible Authentication Strategies: Implemented a new AuthHandler abstraction and a MultiAuthHandler composite, allowing the API to support multiple authentication methods (e.g., OAuth2, API tokens, or no-auth for development) seamlessly.
🧪 Enhanced Testing Infrastructure: Added ASGIAdapter to enable direct in-memory testing of HTTP requests within the ASGI application, significantly enhancing test reliability and performance.
📄 Dedicated API Configuration: Introduced ApiConfig to centralize API-specific settings, separating them from general Azure infrastructure configurations for clearer management.
🛠️ Developer User Information Provider: Added DevUserInformationProvider to simplify local development and testing by providing a configurable user identity without requiring external authentication services.

Changed

🔄 Extended Agent Discovery: Agent discovery responses now include stop_events alongside start_events, providing a more complete picture of an agent's lifecycle and supported termination events. The corresponding StartEventSpecs type was renamed to EventSpecs for broader applicability.
🗓️ Thread Creation Timestamp: Thread entities now automatically record their created_at timestamp upon creation, improving traceability and data management.
⚙️ Updated CI/CD Workflows: Adjusted backend linting and testing actions, including the aihub_api Docker Compose configuration and Poetry lock command, for improved reliability and efficiency in continuous integration.

Fixed

🐛 Refined Azure User Information Error Handling: Improved error handling in AzureUserInformationProvider to raise a ValueError for failed user information fetches from Microsoft Graph, providing more precise feedback.
🐞 Corrected Thread User ID Access: Addressed an issue where thread access checks incorrectly used user_id instead of id for user validation within thread operations.

Security

🔑 Enhanced Agent Access Control: The AuthenticatedUser now properly recognizes the AllAgents role, providing more flexible and robust permissions for agent interactions.
🔒 Authenticated Health Checks: The /health endpoint now requires proper authentication, preventing unauthorized access to application status information and ensuring a more secure deployment.
🛡️ Agent Access Validation for Chat: Chat completion endpoints now perform explicit authorization checks, ensuring users only communicate with agents they are permitted to access.

Refactor

🧹 Streamlined Authentication Modules: Restructured and consolidated authentication-related modules, moving core components to aihub_lib and adopting a unified AuthHandler interface across the application. This includes the removal of legacy use_oauth2_user and use_no_auth_user dependencies.
🗑️ Removed Legacy AccessToken System: The outdated AccessToken persistence entity and its associated code have been removed in favor of the new, more flexible BearerToken system.
📦 Centralized API Testing: Migrated various .http integration test files to comprehensive Python-based pytest suites, improving test maintainability, automation, and reliability.
🧹 Simplified MongoDB Document Access: Refactored RefDoc querying methods to remove the organization_shortname parameter, streamlining data access and consistency.
⚙️ Improved Configuration Clarity: Renamed BaseConfig to AzureBaseConfig and adjusted its usage to better reflect its Azure-specific nature, separating it from general ApiConfig.
🧹 General Code Cleanup: Removed extraneous SonarLint settings from project files and performed minor import reordering for better code hygiene.

[v0.52.0] - 2025-02-19 - Bot Reliability and Code Clarity Improvements

Fixed

🐛 Improved Conversation Message Retrieval: Enhanced the conversation message retrieval mechanism to consistently return a list, ensuring more robust handling of conversation history within the bot service.

Refactor

🧹 Refined Conversation Entity Type Hinting: Updated internal type hints for conversation message access, providing a more accurate representation of the data structure and improving code clarity.

[v0.51.0] - 2025-02-19 - Unveiling the Future of AI-Hub: Tiered Services and Enhanced Architectural Clarity

Added

✨ Introduced Multi-Tiered AI-Hub Offering: Detailed the new tiered approach to AI-Hub services, encompassing Basic LLM Access, Basic+ Integrations, Custom AI Assistants, and Autonomous AI Agents.
🦾 Dedicated Sections for AI Capabilities: Added comprehensive explanations for LLM Access, AI Assistants, and AI Agents, outlining their distinct roles, use cases, and benefits within the AI-Hub ecosystem.
📈 Formalized New Architectural Components: Explicitly documented Bots module (for communication channel integration), Chat UI, Agents Transparency Frontend, and Process Panel as core components of the AI-Hub architecture.

Changed

🔄 Revamped AI-Hub Introduction: The foundational documentation for AI-Hub has been significantly rewritten to reflect its evolution into a comprehensive, multi-tiered platform.
📄 Updated High-Level Architecture Diagram: The architectural overview diagram now clearly illustrates the expanded ecosystem of AI-Hub components, including new user interfaces and integration points.

[v0.50.0] - 2025-02-19 - Smarter Slack Bot Responses in Channels

Changed

🤖 Improved Slack Channel Interaction: Bots, including both Agent and OpenAI variants, now intelligently respond in Slack channels only when explicitly mentioned. This change significantly enhances bot etiquette by preventing unnecessary responses in shared conversations.

Added

✨ Introduced Slack Interaction Utilities: New helper functions, is_slack_channel_message and is_bot_mentioned, have been added to accurately detect Slack channel messages and identify when the bot is mentioned. These utilities are crucial for enabling the new, more selective response behavior in Slack.

[v0.49.0] - 2025-02-19 - Enhanced Bot Architecture and Flexible Deployments

Added

✨ Configurable System Messages: Bots can now be configured with specific system messages directly from the database, allowing for more tailored and dynamic bot behaviors.
💬 Improved Slack Integration: Enhanced handling of Slack threads, ensuring bots respond in the correct context and fetch full conversation history from channels.
🚀 Centralized Streaming Responses: A new common service now handles streaming responses for all bot types, leading to more consistent and efficient real-time user experiences.
💾 Robust Conversation Persistence: Conversation history management has been improved to support bulk message additions and automatically create conversations if they don't exist.
🛠️ MongoDB Setup Option: The Azure Bot setup script now explicitly supports direct MongoDB connection strings, offering more deployment flexibility.
⚙️ Flexible Azure Bot Configuration: The setup_azure_bot.py script now allows for more granular configuration of API URLs and paths, and supports specifying the Azure Bot SKU.

Changed

🔄 Stricter Conversation Schema: The conversation entity schema has been made stricter (strict=True), ensuring data integrity and consistency in stored conversations.
⚡️ Enhanced Azure Bot Setup: The setup_azure_bot.py script now includes pre-emptive deletion of existing bots and improved error handling for a smoother deployment process.

Refactor

🧹 Unified Bot Services: Major internal restructuring of bot logic, moving common functionalities like conversation management, system message retrieval, and activity processing into a shared Service base class.
🦾 Specialized Bot Implementations: Old generic chat bots have been replaced with specialized AgentChatBot and OpenaiChatBot classes (and their streaming counterparts), improving modularity and clarity.
📦 Reorganized Bot Modules: Bot and route modules have been reorganized and renamed to reflect the new, more logical architecture, improving code maintainability.
📖 Simplified Conversation Entity: The ConversationEntity has been streamlined by removing user-specific tracking, focusing solely on message history tied to a conversation ID.

Removed

🗑️ Deprecated Chat Bots: The generic ChatBot and EchoBot implementations have been removed, superseded by the new specialized bot architecture.
🚫 User Tracking in Conversations: Direct user tracking fields have been removed from the ConversationEntity to simplify the database schema and conversation management.

[v0.48.0] - 2025-02-19 - Enhanced Test Stability and Build System Updates

Changed

🚀 Updated the SonarCloud scan GitHub Action to v5, improving the CI/CD pipeline's analysis capabilities and ensuring compatibility with the latest SonarCloud features.

Fixed

🐛 Improved chatbot test stability by significantly extending timeouts for asynchronous operations, which helps prevent flaky test failures related to health checks and streaming response validations.

Added

🔗 Expanded Azure SDK Support: Integrated new Azure SDK dependencies, including azure-mgmt-cognitiveservices, azure-storage-file-datalake, and adlfs, to support enhanced Azure service interactions and data lake functionalities.
🧪 CI/CD Azure Credential Injection: Enabled the injection of AZURE_SUBSCRIPTION_ID into the CI/CD test environment, ensuring robust authentication for automated testing of Azure-dependent components.

Changed

🚀 Azure Subscription Identification: Transitioned from referencing Azure subscriptions by display name to using the direct Azure Subscription ID (GUID), enhancing precision and reliability in resource management.
⚙️ Standardized Azure Resource Naming: Implemented a simplified naming convention for various Azure resources (e.g., Cognitive Search, Document Intelligence, Speech Service, Cosmos DB, and Data Lake), removing environment-specific suffixes for a cleaner and more consistent infrastructure.
✨ Optimized Azure Service Authentication: Streamlined the authentication process for Azure management clients by directly utilizing the Azure Subscription ID, which removes the need for redundant subscription lookups.

Refactor

🧹 Module Structure Modernization: Performed internal refactoring of module structures and import paths (e.g., Configs to configs and vector_stores to stores) to improve code organization and maintainability.
🔄 Cosmos DB Access Consolidation: Replaced the CosmosConnectionStringSingleton with CosmosAccess for a more consistent approach to managing Cosmos DB connections.
🗑️ Configuration Cleanup: Removed the ENVIRONMENT variable from BaseConfig as part of the broader effort to streamline Azure infrastructure configuration.

[v0.46.0] - 2025-02-18 - Streamlined Azure Integration and Configuration

Changed

🔄 Azure Subscription Identification: Transitioned all Azure infrastructure access from using Azure Subscription Names to more robust and direct Azure Subscription IDs, enhancing reliability and security.
⚡️ Azure Resource Naming Convention: Simplified the naming convention for Azure resources by removing the environment-specific slug, promoting a cleaner and more consistent resource management approach.

Refactor

🧹 Configuration Base Class: Restructured the BaseConfig by removing the explicit ENVIRONMENT field, streamlining the configuration process and abstracting environment specificity across the platform.
⚙️ Azure Service Access Logic: Refactored the internal logic for accessing various Azure services (AI Search, Cognitive Services, Cosmos DB, Data Lake) to directly utilize Azure Subscription IDs, improving efficiency and removing reliance on potentially ambiguous subscription name lookups.
🗂️ Internal Module Naming: Standardized internal module naming conventions for configuration files from Configs to configs (lowercase) for improved consistency within the codebase.

Added

⬆️ Azure Management Dependencies: Incorporated new azure-mgmt-cognitiveservices, azure-storage-file-datalake, and adlfs dependencies to support the updated Azure integration patterns and access methods.
🔐 CI/CD Azure Subscription ID Support: Updated CI/CD workflows to pass the AZURE_SUBSCRIPTION_ID secret, ensuring alignment of testing environments with the new Azure configuration standards.

[v0.45.0] - 2025-02-17 - Internal Workflow Enhancements

Changed

🔄 Updated CI/CD Dependency Locking: The GitHub Actions workflow now utilizes poetry lock instead of poetry lock --no-update for dependency management, ensuring more current dependency resolution during build processes.

[v0.44.0] - 2025-02-17 - Flexible Azure Bot Deployments and Development Enhancements

Added

🦾 Introduced Single-Tenant Azure Bot Support: The setup script now allows for the creation and configuration of Azure AD application registrations and Azure Bot resources specific to a single tenant, providing greater deployment flexibility.
⚡️ Enabled Azure OpenAI Chatbot in Playground: The development playground can now run and test OpenAI chat bots directly integrated with Azure OpenAI services, facilitating easier development and experimentation.
📄 Expanded Credential Storage Schema: New fields, including APP_TYPE and APP_TENANTID, have been added to the Credentials entity within PathEntity to support more diverse authentication types and tenant-specific configurations for bots.

Changed

🔄 Dynamic Bot Adapter Retrieval: The Service.get_adapter method now directly accepts the FastAPI Request object, enabling more contextual and robust retrieval of the appropriate bot adapter.
🔑 Updated Credential Field Names: The PathEntity for credentials now uses APP_ID and APP_PASSWORD instead of the previous app_id and app_password, aligning with the new tenant-aware configuration.
🛠️ Enhanced Azure Bot Setup Script: The setup_azure_bot.py script has been significantly updated to manage single/multi-tenant bot configurations, including app registration logic and streamlined Cosmos DB credential storage.

Refactor

🧹 FastAPI Import Consolidation: Standardized the import of Request and Response objects to use fastapi directly across all controllers, improving code consistency.
🗑️ Streamlined Cosmos DB Configuration: Removed the hardcoded Cosmos DB connection string from the development .env file, promoting more secure and dynamic environment variable management.
⚙️ Optimized Cosmos DB ID Generation: The setup_azure_bot.py script no longer explicitly generates object IDs for Cosmos DB entries, allowing the database to handle ID assignment automatically.

[v0.43.0] - 2025-02-17 - Core Tooling and Build System Enhancements

Changed

🚀 Updated Poetry Version: Upgraded the Poetry dependency management tool in GitHub Actions workflows from v1.8.3 to v2.1.1, ensuring compatibility with the latest features and improved build processes.
⚙️ Refined Poetry Core Requirements: Standardized and updated the poetry-core build system requirements across all pyproject.toml files to >=2.0.0,<3.0.0, enhancing build stability and future compatibility.

Refactor

🧹 Updated Node Post-processing: Renamed the process method to postprocess_nodes within ScoreScalerPostProcessor for better clarity and consistency in node manipulation utilities.

[v0.41.0] - 2025-02-17 - Enhanced Azure AI Search Capabilities

Added

✨ Expanded Azure AI Search Configuration: Introduced the ability to specify a semantic_configuration_name when creating Azure AI Search vector stores, providing more control over semantic search behavior and relevance tuning.

[v0.40.0] - 2025-02-14 - Enhanced Azure Bot Management and Multi-Tenancy

Added

✨ Dynamic Bot Credential Storage: Introduced a new PathEntity in aihub_bot.persistence.entities to securely store Azure Bot credentials, enabling flexible per-path bot configurations.
⚙️ Azure Bot Setup Utility: A new setup_azure_bot.py script simplifies the setup process for Azure Bots, automating Azure AD app registration, credential management, and their storage within Cosmos DB.

Changed

🔄 Multi-Tenancy for Azure Bots: The aihub_bot framework now supports running multiple Azure Bots from a single application instance, each with its own credentials and accessible via distinct API paths.
🚀 Dynamic Adapter Management: The Service class no longer uses a static global adapter, instead providing a get_adapter method to dynamically create CloudAdapter instances based on the request path, retrieving corresponding credentials from the new PathEntity.
⚡️ Controller Response Types: Bot controllers now return Response objects instead of JSONResponse for broader compatibility in processing bot activities.

Refactor

🧹 Persistence Module Restructure: The aihub_bot persistence entities have been reorganized by moving ConversationEntity and associated files to a more general aihub_bot.persistence.entities module.
🧹 Centralized Bot Configuration: Removed static Azure Bot credentials (APP_ID, APP_PASSWORD, APP_TENANTID) from aihub_lib.infrastructure.azure.BaseConfig, as these are now dynamically managed via the PathEntity for enhanced security and flexibility.
🧹 Streamlined Service Class: The on_error handler and static ADAPTER have been removed from the general Service class, contributing to a cleaner and more focused design for bot service routing.

[v0.39.0] - 2025-02-14 - Improved RAG Agent Rejection Handling and Data Consistency

Changed

💬 Enhanced RAG Agent Rejection Responses: The RAGAgent now includes the limited chat history (without context) when generating responses for rejected queries. This provides the Large Language Model (LLM) with more comprehensive context to explain the reasons for rejection, leading to more informative and relevant user messages.

Refactor

⚙️ Standardized Timestamp Processing: The internal utility for formatting Unix timestamps has been updated to explicitly expect and handle integer timestamps, improving the robustness and consistency of date-time metadata processing across the system. This refactor includes corresponding updates to test cases to align with the new integer type expectation for metadata fields such as inserted_at, updated_at, and created_at.

[v0.38.0] - 2025-02-13 - Bot Stability and OpenAI Integration Enhancements

Fixed

🐛 Improved Robustness for Chat Roles: Ensured that chat message roles default to "user" if not explicitly provided, preventing potential issues with None values in bot interactions.
⚡️ Enhanced OpenAI Streaming Reliability: Added robust error handling for OpenAI chat completions, ensuring the stream continues smoothly even if content chunks are empty or null.

[v0.37.0] - 2025-02-13 - Enhanced RAG Context and Expanded Localization

Added

✨ Expanded Localization Support: Introduced French and Italian translations for agent rejection messages and agent thoughts, improving multi-lingual capabilities.
⚙️ Configurable Locale Paths: Added the ability to specify locale paths dynamically through agent configuration, increasing flexibility for internationalization.
📄 Rich RAG Context Metadata: Enhanced the format of RAG context documents to include comprehensive metadata (e.g., source, namespace, language, version, timestamps), providing more detailed context to the LLM.
💡 New Default RAG Context Prompt: Introduced a standardized, detailed prompt for RAG agents within the aihub_lib for consistent context presentation.

Changed

🔄 Centralized Core Prompts: Moved condenser and RAG-related context prompts from individual agents to the aihub_lib for better reusability and maintainability.
📝 Improved Condenser Prompt Wording: Refined the language of the condenser.standalone_question prompt across all supported locales for greater clarity and accuracy.

Removed

🗑️ Redundant Agent Prompts: Removed specific condenser and rag_agent prompts from aihub_agent's internal translations, as they are now managed centrally in aihub_lib.
🗑️ Outdated Test Prompts: Removed test prompts from agent translation files in aihub_agent.

[v0.36.0] - 2025-02-13 - Streamlined Configuration & Azure Bot Service Enhancements

Added

⚙️ Extended Azure Bot Service Configuration: Introduced new APP_TYPE and APP_TENANTID fields within the BaseConfig to provide more comprehensive configuration options for Azure Bot Service deployments.

Changed

⚡️ Improved Streaming Chat Completion Responsiveness: Enhanced the OpenaiChatService to ensure smoother and more responsive streaming of chat completions by optimizing event loop handling.
🔄 Streamlined Self-Hosted LLM Configuration: The SelfHostedLLMConfig now automatically initializes its default_parameter if not explicitly provided, simplifying model setup and reducing boilerplate for developers.
🧪 Updated Playground LLM Configurations: The main.py playground example in aihub_bot has been updated to reflect the streamlined LLM configuration and now includes gpt-4o-mini as an example Azure OpenAI model.

[v0.35.0] - 2025-02-12 - Agent Intelligence Boost: Few-Shot Guarding and Workflow Optimization

Added

✨ Few-Shot Guarding for RAGAgent: Introduced a new mechanism to filter user queries based on predefined few-shot examples. This allows RAG agents to explicitly accept or reject requests with specific reasons, significantly enhancing control over agent scope. New configuration options (few_shot_guard_examples) and events (FewShotAcceptEvent, FewShotRejectEvent) support this feature.
🦾 Direct User Query Access: Added a convenient user_query property to the StartEvent, providing a straightforward way to access the last user message sent, simplifying agent workflow logic.
📄 Localized Few-Shot Guard Prompts: Included comprehensive internationalized prompts for the new few-shot guard, ensuring consistent and flexible agent behavior across various locales.
🚀 Rejected Query Responses: Implemented logic within the RAGAgent to automatically provide the user with a specific, configurable reason when a query is rejected by the few-shot guard.

Changed

🔄 RAG Agent Workflow Integration: Modified the RAGAgent's processing flow to seamlessly integrate the new few-shot guard. This ensures that knowledge retrieval and subsequent responses are contingent on the initial query's acceptance.
⚙️ Optional RAGAgent Prompts: Made condense_question_prompt and context_prompt optional in RAGAgentConfig, offering greater flexibility for agent configuration and allowing default prompts to be used if not specified.
🧹 Improved Standalone Question Condensation: Enhanced the condense_standalone_question utility to explicitly filter out system messages from the chat history, leading to more accurate and focused standalone question generation.

Refactor

📦 Centralized Agent Utilities: Reorganized common agent events and configurations (like LimitChatHistoryEvent and StandaloneQuestionCondenserEvent) by moving them into a new aihub_agent/agents/common directory, improving modularity and reusability across different agent types.
🔗 Decoupled Agent Logic from RunContext: Refactored both FewShotAgent and RAGAgent to directly pass essential data (such as user query and chat history) via event objects. This reduces reliance on the RunContext for state management, simplifying method signatures and improving code clarity.
🌎 Moved Condenser Prompt Translations: Relocated the standalone question condenser prompt translations from aihub_agent to aihub_lib, centralizing common prompt templates within the core library for better management.
📁 Standardized Directory Naming: Renamed Configs to configs and Events to events within the rag agent modules for consistent lowercase naming conventions across the project.

[v0.34.0] - 2025-02-11 - Introducing `aihub_bot` and Core Project Enhancements

Added

✨ Introduced aihub_bot module: A new core module dedicated to bot functionalities, enabling new conversational AI capabilities within the platform.

Refactor

🧹 Refined OpenAI Chat Service API: The internal method for streaming OpenAI chat completions within StreamOpenaiChatBot has been renamed to stream_on_message_completion for better clarity and consistency with OpenAI's API.
⚙️ Streamlined Project Configuration: Updated internal IDE configuration files across all modules to improve project structure definition and ensure a consistent development environment setup.

Removed

🗑️ Removed Playground Environment Configuration: The dedicated .env file for the bot playground testing has been deleted, streamlining local environment management.

[v0.33.0] - 2025-02-10 - Enhanced RAG Agent with Contextual Node Retrieval

Added

✨ Introduced Contextual Node Retrieval for RAG Agent: The RAG agent now supports fetching adjacent (previous and next) nodes from the vector store based on the initial retrieved results, significantly enhancing the contextual understanding for generated responses.
⚙️ New RetrievePrevNextConfig Option: A new configuration option has been added to the RAG agent's RetrieveStepConfig, allowing users to control the behavior of contextual node retrieval, including the number of nodes and traversal direction (previous, next, or both).
🛠️ Developed VectorPrevNextPostProcessor: A new internal post-processor and associated utility functions were introduced to efficiently manage the traversal and retrieval of related nodes in vector stores.
🧪 Expanded CI/CD for aihub_lib: The CI/CD workflow for the aihub_lib module now includes Milvus standalone docker-compose, enabling more robust and comprehensive testing for vector store-related functionalities.

Refactor

🧹 Centralized Milvus Test Utilities: The milvus_vector_store_content module, containing utilities for Milvus vector store content management, has been moved from the RAG agent's playground to aihub_lib/aihub_lib/testing to improve organization and reusability across the platform.

[v0.32.0] - 2025-02-10 - Expanded Bot Capabilities: OpenAI Integration and Refined Agent Interactions

Added

✨ OpenAI Chat Integration: Introduced new controllers and services to allow direct chat interactions with OpenAI and Azure OpenAI models, supporting both JSON and streaming responses.
🚀 Centralized Bot Authentication: Added APP_ID and APP_PASSWORD fields to BaseConfig in aihub_lib, centralizing Azure Bot Service authentication configuration.

Refactor

🧹 Agent Chat Module Reorganization: Extracted and moved agent-specific chat logic into a dedicated aihub_bot/bots/chat/agent and aihub_bot/routes/chat/agent package structure, including JsonAgentChatBot, StreamAgentChatBot, and AgentChatService for clearer separation of concerns.
🔄 Generalized ChatBot: The ChatBot class in aihub_bot has been refactored to serve as a more generic base class, removing agent-specific dependencies from its core.
⚡️ Consolidated Bot Configuration: Replaced module-specific DefaultConfig with the centralized BaseConfig from aihub_lib in the aihub_bot module, streamlining configuration management.

Changed

↔️ Agent Chat Endpoint Paths: The API endpoints for agent-based chat completions have been updated from /chat to /agent/chat to reflect the new modular structure.
⏳ Stream Response Timeout: Increased the timeout for receiving chunks in stream responses from 0.5 to 1 second for improved robustness.

Fixed

🐛 Robust Conversation Creation: Ensured that a new conversation is automatically created if it doesn't exist when adding a message, preventing potential errors during the first interaction.
🤖 Improved Bot Message Role Assignment: Enhanced the logic for assigning the 'bot' role to outgoing messages to ensure consistency.
⚙️ Corrected Activity Model Type Mapping: Adjusted the internal type mapping for object in activity models for improved data handling.

[v0.31.0] - 2025-02-07 - Enhanced Tracing Authentication

Added

✨ Introduced Phoenix Tracing Authentication: Added support for configuring and using a PHOENIX_AUTH_TOKEN to authenticate tracing data sent to the Phoenix endpoint, enhancing the security and control over your observability data.

[v0.30.0] - 2025-02-07 - Improved Chat Event Handling

Changed

🔄 Streamlined User Message Event Creation: Simplified the construction of UserMessageEvent within the ChatService by directly providing the full list of messages, enhancing consistency and streamlining how user message events are handled.

[v0.29.0] - 2025-02-07 - Enhanced Bot Connectivity and Development Workflow

Changed

⚡️ Improved NATS Connectivity for Bots: The aihub_bot now dynamically retrieves its NATS endpoint from NatsConfig, enhancing deployment flexibility and making the connection environment-agnostic.
📄 Updated Semantic PR Workflow: Added bots as a recognized type in the semantic PR workflow, standardizing and streamlining the development process for bot-related contributions.

[v0.28.0] - 2025-02-07 - OpenAI API Emulation and Unified AI Resource Management

Added

🚀 OpenAI API Emulation: Introduced a comprehensive OpenAI API-compatible endpoint suite, enabling seamless integration with existing OpenAI SDKs for various generative AI tasks. This includes:
- 🤖 Chat Completions: Interact with configured chat models using the /openai/chat/completions endpoint, supporting both JSON and streaming responses.
- 🔍 Embeddings: Generate text embeddings via /openai/embeddings for various models.
- 🖼️ Image Generation: Create images using models like DALL-E through the /openai/images/generations endpoint.
- 🎤 Speech-to-Text (STT): Transcribe audio files to text using /openai/audio/transcriptions.
- 🗣️ Text-to-Speech (TTS): Convert text into natural-sounding speech through /openai/audio/speech.
- 📄 Model Management: List and retrieve details for available models via /openai/models and /openai/models/{model_name}.
✨ New AI Resource Configurations: Expanded aihub_lib with dedicated configuration classes for different generative AI resource types:
- ImageModelConfig and AzureOpenaiImageModelConfig for image generation models.
- STTConfig and AzureOpenaiSTTConfig for speech-to-text models.
- TTSConfig and AzureOpenaiTTSConfig for text-to-speech models.
⚙️ Open WebUI Docker Compose Example: Provided a docker-compose-open-webui.yml example to demonstrate integration with Open WebUI, leveraging the new OpenAI API emulation.

Refactor

🔄 Unified Generative AI Resource Structure: Performed a significant refactoring of aihub_lib.generative_ai.llms into aihub_lib.generative_ai.resources, introducing a more generic and extensible structure for managing all AI model types (LLMs, Embeddings, Images, STT, TTS).
- 🧹 Centralized Base Configurations: Introduced ResourceConfig as the new base class for all AI resource configurations, along with AzureOpenaiResourceConfig for Azure-specific models, promoting consistent configuration patterns.
- 🧹 Consistent Parameter Naming: Standardized the api_endpoint parameter to base_url across all model configurations for clarity and consistency.
- 🧹 Parameter Class Renaming: Renamed ModelParameter to LLMModelParameter (and its specialized versions like ChatLLMParameter and EmbeddingLLMParameter) to better align with the new resource hierarchy.

Changed

🦾 Agent Configuration Updates: Updated FewShotAgent, LLMWrappingAgent, and RAGAgent configurations to utilize the new aihub_lib.generative_ai.resources paths and the base_url parameter, reflecting the core library's refactor.
💬 Chat Service Updates: Adjusted the internal ChatService in aihub_api to align with the new generative AI resource paths.

[v0.27.0] - 2025-02-06 - Introducing Chat Streaming and Enhanced API Definitions

Added

✨ Asynchronous Chat Streaming: Implemented new capabilities for streaming chat responses from AI agents to users, providing real-time updates as content is generated. This allows for a more dynamic and responsive user experience.
📄 Dynamic OpenAPI Schema Generation: Introduced a new model to dynamically generate OpenAPI schemas for Activity payloads, significantly improving API documentation and ensuring robust request validation for bot interactions.
⚡️ Streaming Endpoint for Agent Interactions: A new /completions/{agent_class}/{agent_id}/stream endpoint was added to support real-time, chunked responses from AI agents, specifically designed for streaming interactions.

Changed

🚀 Synchronous Chat Endpoint Updates: The existing synchronous chat endpoint (/completions/{agent_class}/{agent_id}/json) now leverages the refactored JsonChatBot and includes enhanced OpenAPI documentation for better clarity and consistency.

Refactor

🧹 Bot Logic Decoupling: Refactored the core ChatBot by extracting specific message handling logic into dedicated JsonChatBot and StreamChatBot classes. This enhances modularity, improves maintainability, and allows for distinct chat interaction patterns (e.g., JSON vs. streaming).
🔄 Bot Service Endpoint Update Handling: Updated the test runner and the main bot service endpoint to correctly process PUT requests, which is essential for updating message activities during streamed responses to users.
🧹 Minor Code Enhancements: Applied @override decorators for clarity and adherence to best practices, along with minor formatting adjustments.

[v0.26.0] - 2025-02-06 - Robust Event Management and Enhanced Data Persistence

Added

✨ Enabled Recursive Event Serialization: The BaseEvent model_dump method now intelligently serializes nested BaseEvent instances, ensuring comprehensive and accurate data persistence for complex event structures.

Changed

🔄 Enhanced Event Deserialization: The DistributedEventStore has been updated to fetch and deserialize events more flexibly by using event class names directly and leveraging a generic ControlEvent.deserialize_event method. This improves support for polymorphic event types and simplifies retrieval logic.

Fixed

🐛 Improved Error Handling in Event Store: Refined error handling within the DistributedEventStore to specifically catch KeyNotFoundError when retrieving events and added comprehensive logging for other exceptions, enhancing the store's reliability and diagnostic capabilities.

[v0.25.0] - 2025-02-06 - Enhanced Testing Framework

Refactor

🧹 Refactored MultistepHumanInTheLoopAgent Tests: Streamlined the test suite by consolidating multiple BDD-style test steps into a single, cohesive pytest function for improved clarity and maintainability.
🔄 Migrated Test Implementation: Transitioned the MultistepHumanInTheLoopAgent tests from a pytest-bdd driven approach to a direct pytest implementation, enhancing test maintainability and reducing boilerplate.

Removed

🗑️ Deprecated pytest-bdd Feature Files: Removed the dedicated pytest-bdd feature file (multistep_human_in_the_loop_agent.feature) and its associated pytest-bdd imports and decorators from the MultistepHumanInTheLoopAgent test suite.

[v0.24.0] - 2025-02-06 - Core Module Synchronization and CI/CD Enhancements

Changed

✨ Synchronized Core Module Versions: All aihub core modules (aihub_agent, aihub_api, aihub_bot, aihub_lib, aihub_pipeline) and their internal aihub_lib dependencies have been consistently updated to v0.24.0, ensuring full compatibility and alignment across the ecosystem.

Refactor

🧹 Streamlined CI/CD Test Workflow: The GitHub Actions test workflow now references the test_backend action locally within the repository, improving build reliability and simplifying the continuous integration setup.

[v0.23.0] - 2025-02-05 - Empowering Chatbots with New Bot Service and Enhanced Core Reusability

Added

🤖 New AI Hub Bot Service: Introduced a dedicated aihub_bot component for seamless integration with Azure Bot Service, enabling conversational AI capabilities.
💬 Chatbot and Echo Bot Implementations: Shipped initial ChatBot for managing AI agent interactions and an EchoBot for basic conversational responses within the new bot service.
💾 Conversation History Persistence: Implemented ConversationEntity within the bot service to store and retrieve chat history, providing context for ongoing conversations.
🚀 Bot Application Runners and Test Utilities: Added BotRunner, BotTestRunner, and SimulatedAgentBotTestRunner to facilitate streamlined development, testing, and deployment of bot applications.
🌐 Static Frontend for Bot Testing Playground: Included basic static web files to enable local testing and interaction with the new bot service.

Changed

⚙️ Renamed API Chat Endpoints: Renamed chat-related methods in aihub_api (e.g., start_api_stream_chat_interaction) to clarify their role as API-specific entry points, distinguishing them from core logic now in aihub_lib.
🛡️ Improved Thread Authorization Error Handling: Enhanced error messages and consistency for unauthorized thread access and modification in the aihub_api's thread controller.
🛠️ Updated CI/CD and Build Configurations: Modified GitHub Actions workflows and the Makefile to integrate the new aihub_bot component into linting, testing, and deployment processes.
🔄 Dependency Management Script Update: The switch_dependency.py script now supports managing aihub_lib dependencies within the aihub_bot project.

Refactor

📦 Core Component Reusability (aihub_lib): Moved several foundational components (e.g., AuthenticatedUser, Controller, HealthController, NATS dependencies, WebSocket receiver components, and core chat service logic) from aihub_api into the shared aihub_lib. This significantly improves modularity and reduces duplication across services.
🚚 Centralized Docker Compose Files: Moved docker-compose.yml and milvus-standalone-docker-compose.yml to the project root, making them accessible and standard for all microservices.

[v0.22.0] - 2025-02-04 - Internal System Alignment and Workflow Stabilization

Changed

🔄 Updated Core Dependency Versions: Synchronized all core modules (aihub_agent, aihub_api, aihub_pipeline) to utilize the latest v0.22.0 release of aihub_lib, ensuring consistent functionality and compatibility across the system.
⚙️ Refined CI/CD Workflow References: Updated GitHub Actions workflows (analyze, lint, review) to reference the stable main branch of AIHub Core actions, enhancing the reliability and consistency of automated checks.

[v0.21.0] - 2025-02-04 - CI/CD Modernization: Streamlined Workflows with Reusable Actions

Added

🚀 Introduced Reusable GitHub Actions: A new aihub_action directory has been added, containing a suite of modular and reusable GitHub Actions. These actions encapsulate common CI/CD tasks such as backend testing, linting (backend and frontend), SonarCloud scanning, pytest coverage commenting, and PR review, significantly streamlining workflow definitions and promoting reusability.
📄 New Actions Documentation: A README.md file is now available in the aihub_action directory, providing comprehensive documentation for all new reusable GitHub Actions and their usage.
🏷️ Enhanced Pytest Markers: New Pytest markers (azure, self_hosted, slow, integration, experimental) have been added and strict marker usage enforced in aihub_agent and aihub_lib for improved test organization and filtering.

Refactor

🔄 Centralized CI/CD Workflows: The existing .github/workflows have been extensively refactored to leverage the newly introduced reusable GitHub Actions. This significantly reduces redundancy, improves maintainability, and standardizes CI/CD practices across the repository.
🧹 Unified SonarCloud Scans: The SonarCloud scanning process has been unified, integrating the aihub_web scan into the primary sonarcloud-scan job for a more cohesive analysis pipeline.

[v0.20.0] - 2025-02-03 - Introducing Few-Shot Agents and Core System Improvements

Added

🦾 Few-Shot Agent: A new agent type has been introduced, enabling agents to generate responses based on provided examples, significantly enhancing their learning capabilities for specific tasks.
🚧 Agent Description Guard: A new utility has been added that uses an LLM to evaluate if a user's query aligns with an agent's defined purpose, improving intelligent request routing and guardrail mechanisms.
🛑 Guard Rejection Event: A new event, GuardRejectionEvent, is now emitted when an agent's internal guardrail system rejects a request, providing a clear reason for the rejection to the client.
📚 Few-Shot Prompting Utilities: New models and functions (FewShotExample, create_few_shot_messages) have been added to the aihub_lib to standardize and simplify the creation of few-shot messages for LLMs.
🌍 Localized Translation Strings: New localized messages have been added to support the newly introduced Few-Shot Agent and the Agent Description Guard, ensuring multilingual support.
🧪 Few-Shot Agent Playground Examples: Comprehensive examples and tests for the new Few-Shot Agent have been added to the playground, demonstrating its capabilities and usage.

Changed

⚙️ Test Command Filtering: The Makefile in aihub_agent has been updated to exclude Azure-specific tests from the default test command, allowing for more granular control over test execution.
📄 Documentation Clarity: The README.md has been updated to provide clearer guidance on agent creation and testing methodologies, distinguishing between run.py for continuous execution and trigger.py for limited runtime scenarios.
🏷️ RAG Agent Test Categorization: RAG Agent tests are now tagged with @azure and @self_hosted to facilitate better test organization and selective execution in CI/CD pipelines.

Refactor

🔄 Core Library Consolidation: Fundamental components like AgentConfig and EventDisplayer have been relocated from aihub_agent to aihub_lib. This centralizes common functionalities, promotes reusability, and streamlines dependencies across agent and API projects, requiring numerous import path updates.
🧹 Tracing Decorator Removal: The @tracing() decorator has been removed from specific document loaders (DocumentIntelligenceLoader, RawLoader), indicating a shift towards a more integrated or alternative approach to tracing within the library.
🌲 New Package Structures: New package directories have been introduced across aihub_agent and aihub_lib to support a more organized and modular architecture for new features and existing components.
⬆️ Document Intelligence Integration: The DocumentIntelligenceLoader has been updated to align with the latest DocumentContentFormat enum from Azure's SDK, ensuring compatibility with updated dependency APIs.

[v0.19.0] - 2025-01-31 - Core Component Release Synchronization

Changed

🔄 Component Version Alignment: All core components, including aihub_agent, aihub_api, aihub_lib, and aihub_pipeline, have been synchronized and updated to version v0.19.0. This ensures all projects utilize the latest features and improvements from aihub_lib.

[v0.18.0] - 2025-01-31 - Refined Release Workflow

Refactor

🧹 Improved Release Tagging Workflow: Reordered the SSH key cleanup step in the GitHub Actions workflow to ensure it executes consistently after all tagging operations, enhancing the robustness of the release process.

[v0.17.0] - 2025-01-31 - Enhanced Agent Orchestration, Flexible LLM Integration, and Developer Experience Boost

Added

🦾 Agent-in-the-Loop (AITL) Delegation: Implemented a new mechanism for agents to delegate tasks to other specialized agents, enabling complex, collaborative workflows. This includes new event types for requests, responses, and exceptions during delegation.
✨ Self-Hosted LLM and Embedding Support: Introduced configurations and examples for integrating self-hosted Large Language Models (via llama.cpp) and embedding models (via Hugging Face Text Embeddings Inference) into agents and the local development environment.
🚀 Milvus Vector Store Integration: Added support for Milvus as a vector store backend for RAG agents, complete with local Docker Compose setup and utilities for content management.
🖼️ Project-Wide IDE Configuration: Included .idea files for consistent PyCharm project settings and improved developer experience, ensuring these files are automatically ignored by Git.
⚙️ NATS Configuration Management: Introduced a dedicated NatsConfig for centralizing NATS endpoint configuration via environment variables, enhancing flexibility for various deployment environments.
📝 Comprehensive Test Coverage for Core Workflows: Added new BDD test scenarios and corresponding test files for conditional, context, fan-out, multi-locale, and semantic event workflows, ensuring robustness and clarity of agent behaviors.
🔗 to_subject Method for PartialAgentTopic: Added a utility method to PartialAgentTopic for easier generation of NATS subjects from partial topic information.

Changed

🔄 Generalized LLM and Embedding Configurations: Refactored RAGAgentConfig to use more abstract ChatLLMConfig and EmbeddingLLMConfig, allowing for greater flexibility in LLM and embedding model choices beyond Azure OpenAI.
⚡️ Flexible Vector Store Integration in RAG: Updated RetrieveStepConfig to directly accept a generic BasePydanticVectorStore instance, replacing the explicit index_name and allowing seamless integration of various vector database solutions like Milvus or Azure AI Search.
🐳 Improved Docker Compose Setup for Development: Migrated aihub_agent's local Docker setup to docker compose (v2) syntax, replaced Ollama with llama.cpp and HF TEI for self-hosted LLM/embedding, and enhanced health checks for NATS and MongoDB, including new Milvus services.
📄 Standardized Test Coverage in Makefiles: Updated Makefile targets across all services (aihub_agent, aihub_api, aihub_lib, aihub_pipeline) to use --cov=. for more comprehensive and consistent local test coverage reporting.
🌐 API Endpoint Naming Convention: Standardized LLM and embedding model configurations to use base_url instead of api_endpoint for consistency across different LLM providers.
🧩 Recursive Event Deserialization: Enhanced BaseEvent deserialization to correctly handle nested event structures, improving the robustness of event processing.
🚀 Git Auto-Tagging Workflow Refinement: Streamlined the auto-tagging GitHub Actions workflow to trigger only on main branch pushes.
⬆️ CI/CD Model and API Version Updates: Upgraded the review-pr.yml GitHub Actions workflow to use gpt-4o and a newer Azure OpenAI API version (2024-08-01-preview) for PR reviews, alongside updated secret and variable names.
🪢 Agent tokenizer Abstraction: Introduced a tokenizer property in ChatLLMConfig and updated its usage in limit_chat_history_with_context, allowing for more flexible tokenization based on the configured LLM.

Refactor

🧹 Consolidated Playground Workflow Examples: Reorganized and renamed various minimal workflow examples within the playground directory for better categorization and clarity.
💅 Minor Code Formatting and Linting: Applied consistent code formatting and linting rules across the codebase, including the adoption of black for Python code.
🗑️ Removed Redundant clean Makefile Targets: Eliminated duplicate clean targets from individual microservice Makefiles, as general project cleaning is now handled at the root level.

Removed

🗑️ Deprecated ollama Service: Removed the ollama Docker service from the aihub_agent playground, replaced by more specific self-hosted LLM inference solutions (llama.cpp and HF TEI).
🧹 Specific LLM Stop Event: Removed the LLMStopEvent class, streamlining event handling as LLMEvent and StopEvent can be used directly or composed for similar functionality.

[v0.16.0] - 2025-01-22 - Unleashing Self-Hosted LLMs and Bolstering CI/CD

Added

✨ New DisplayingAgent Example: Introduced a new example agent (DisplayingAgent) with comprehensive BDD tests to showcase event display capabilities within agent workflows.
🦙 Ollama Integration for Local Development: Added ollama as a service in the docker-compose.yml for simplified local setup and testing of self-hosted Large Language Models.
⚙️ Flexible Tokenizer Configuration for Self-Hosted LLMs: Enhanced SelfHostedLLMConfig with a new tokenizer_name field, allowing precise specification of the tokenizer for self-hosted models, improving compatibility and performance.
🧪 New BDD Tests for LlamaIndexAgent: Implemented new BDD tests to validate the LlamaIndexAgent's functionality, specifically with self-hosted LLM configurations.
🔍 Direct StartEvent Retrieval: Added a get_start_event method to AgentTestRunner for easier access to the initial start event during agent testing.

Changed

🔄 Expanded LLM Support for LlamaIndexAgent: The LlamaIndexAgent now supports SelfHostedLLMConfig, providing the flexibility to integrate with a wider range of self-hosted Large Language Models alongside Azure OpenAI.
🚀 Improved CI/CD Workflow Reliability: Updates to add-tag.yml and analyze-test-pr.yml workflows, including refined git configurations, explicit user/email settings for bot commits, and improved dependency management, ensuring more robust and consistent automated tagging and testing.
⏱️ Enhanced Test Setup with Ollama Readiness Checks: Added readiness checks for the ollama service in CI/CD tests, ensuring that self-hosted LLM services are fully available before test execution.
📦 Streamlined Global Dependency Management: Introduced steps to validate and install global poetry dependencies and use local core, making the CI/CD test environment more consistent and efficient.

Refactor

🧹 Monorepo Project Structure: Adopted a global pyproject.toml at the root for better management of the monorepo, simplifying project-wide dependency and tool configurations.
🗑️ Removed start_service.sh: The dedicated start_service.sh script has been removed, indicating a shift towards more standardized poetry run commands for service startup and local development.
🛠️ Refined CI/CD Script Logic: Updated the method for iterating over directories and reordered the tag deletion process within the auto-tagging workflow for improved script robustness and clarity.

Changed

⚙️ CI/CD Workflow Stability: Enhanced the Pytest coverage comment job within pull requests to ensure it exclusively triggers on pull request events, improving workflow reliability.

[v0.14.0] - 2025-01-21 - Streamlined Workflows and Robust Testing

Added

🚀 Automated Draft PR Creation: Introduced a new GitHub Actions workflow that automatically creates a draft pull request when a new branch is pushed, streamlining early-stage development.
🧪 Comprehensive Module Testing: Implemented granular make test-cov commands and integrated pytest-cov across aihub_agent, aihub_api, aihub_lib, and aihub_pipeline, enabling detailed test coverage reporting for each module.
📊 Automated Pytest Coverage Comments: Added a new GitHub Actions job to automatically publish Pytest and coverage results as comments on pull requests, providing immediate feedback on test status.
🔎 Generalized SonarCloud Scans: Refactored SonarCloud analysis workflows to use a matrix strategy for all Python modules, streamlining code quality checks across the entire codebase.
🌐 Frontend Linting Workflow: Introduced a dedicated GitHub Actions workflow for linting the aihub_web (frontend) code using ESLint, ensuring consistent code style and quality.
📚 Package Management Documentation: Added a comprehensive "Package Management" section to the README.md, detailing the multi-package structure, versioning, and internal dependency referencing.
✨ Agent Discovery Test Enhancements: Expanded AgentTestRunner with new properties and a wait_for_event method to facilitate robust testing of agent discovery events and asynchronous workflows.
⚙️ Enhanced Agent Workflow Examples: Introduced new and expanded BDD test scenarios for configured, discoverable, human-in-the-loop, and optional workflow agents, showcasing more complex and configurable agent interactions.
🌍 Multilingual Support Placeholders: Added new empty translation files as placeholders for future French and Italian language support.

Changed

🔄 Internal Dependency Management: Switched internal module dependencies (e.g., aihub_agent depending on aihub_lib) from local file paths to Git tags, enhancing consistency and simplifying deployments.
⚙️ CI/CD Trigger Conditions: Adjusted various GitHub Actions workflows (AI Code Review, Semantic PR checks) to prevent execution on draft pull requests, reducing unnecessary CI runs.
🌐 Nuxt.js Configuration: Updated the Nuxt.js configuration for the frontend, including enabling SSR (ssr: false) and integrating @nuxt/eslint for improved linting.
🧹 LocaleHandler Usage: Modified LocaleHandler in aihub_lib to be instantiated for better object-oriented practice in translation handling.
📄 switch_dependency.py Script: Improved the switch_dependency.py script by adding develop: True for local path dependencies, streamlining local development setup.

Fixed

🐛 Storybook Path Correction: Corrected relative paths in aihub_web/.storybook/main.ts from ..\stories to ../stories, resolving potential issues with Storybook loading.

Refactor

♻️ CI/CD Version Bumping: Overhauled the add-tag.yml workflow to generalize the version bumping process for pyproject.toml files across multiple directories and update the Makefile tag, making releases more reliable.
🏗️ Centralized Test & Analyze Workflow: Consolidated multiple SonarCloud jobs into a unified test-modules and sonarcloud-scan workflow, improving efficiency and maintainability of the CI/CD pipeline.
🧹 Backend Linting Generalization: Refactored the backend linting workflow to apply black and isort across all Python modules using a matrix strategy, ensuring consistent code style.
📂 Internal Module Path Restructuring: Moved node metadata constants from aihub_lib.constants to aihub_lib.persistence.rag.vectors.node_metadata, improving logical organization within the library.
⚙️ Frontend Build System: Migrated tailwind.config.js to tailwind.config.mjs and updated import statements for ES module compatibility in the frontend.
💡 Playground Workflow Renaming: Renamed discovery_workflow to discoverable_workflow for clearer semantic meaning, alongside renaming some associated test files and agent types.

[v0.13.0] - 2025-01-17 - Enhanced Workflow Observability and Type System Robustness

Added

✨ Enhanced Step Debugging Capabilities: Introduced comprehensive debug logging for workflow steps, providing more detailed insights into their configuration and event mappings for easier troubleshooting.
⚡️ Enabled Logging in Playground Example: Activated logging within a minimal workflow example to assist with debugging and understanding runtime behavior.

Refactor

🧹 Improved Type Extraction for Fixed-Size Lists: Refactored the internal mechanism for extracting item types from FixedList annotations, leading to more robust and explicit type handling within the workflow system.

[v0.12.0] - 2025-01-17 - Streamlined Development Environment & Setup

Changed

🔄 Improved PyCharm Development Setup: The README.md now provides revised and clearer instructions for setting up multi-microservice projects in PyCharm. The new recommended approach leverages PyCharm's "Attach Project" feature, allowing developers to manage all aihub-core microservices within a single PyCharm window while maintaining separate and isolated Poetry environments for each service, significantly enhancing the developer experience for complex projects.

Removed

🗑️ Obsolete PyCharm Run Configurations: Removed outdated PyCharm run configuration files (.run/*.xml) that are no longer necessary following the revised development environment setup instructions.

[v0.11.0] - 2025-01-16 - Smart Agents Get Smarter: RAG Capabilities & Enhanced Developer Experience

Added

✨ Introduced RAG Agent: A new, full-fledged Retrieval-Augmented Generation (RAG) agent, RAGAgent, is now available, enabling knowledge-aware conversational AI.
🛠️ RAG Utilities: A suite of new utilities has been added for RAG workflows, including functions to combine nodes in order, condense standalone questions, limit chat history with context, and retrieve nodes, facilitating advanced contextual processing.
📊 Score Scaling Post-processor: A ScoreScalerPostProcessor has been introduced to normalize retrieval scores, improving the relevance ranking of retrieved information.
📖 Agent Development Guide: A comprehensive README.md for aihub_agent now provides detailed guidance on developing, testing, and documenting new AI agents.
🗣️ Localized Prompts & Thoughts: New localized prompt and thought messages have been added for the RAG agent, improving multi-language support and user experience.

Changed

🧹 Code Quality Enforcement: Integrated ruff, isort, and black configurations into aihub_agent and aihub_api projects, along with a new pr-ready Makefile target, to ensure consistent code style and quality.
🏷️ Enhanced Step Metadata: The @step decorator now supports name and description parameters, allowing for more detailed and localized metadata for workflow steps.
📈 Improved Workflow Visualization: The NetworkXVisualizer has been enhanced to leverage new step metadata, providing richer and more informative workflow diagrams.
🔗 Event System Refinements: Refined event handling logic and structure across various nats.events classes, including additions for Document and RetrieverEvent to support advanced data flow in RAG workflows.

Refactor

🧪 Refactored AgentTestRunner: The AgentTestRunner has been refactored to improve test environment control, including new context managers for streamlined agent testing.
⚙️ Internal Build Process Updates: Updated internal dependency references and Makefile structures to streamline build and test processes for aihub_agent and aihub_api.

[v0.10.0] - 2025-01-16 - Introducing Robust Branch Protection Policies

Added

🛡️ Implemented Comprehensive Branch Protection Policies: New documentation has been added to the README, detailing robust branch protection rules for main and initiative branches. These rules enforce restricted direct pushes, prevent deletions and force pushes, require a linear commit history, and mandate pull requests with specific approval processes (including squashed merges), significantly enhancing codebase stability and integrity.

[v0.9.0] - 2025-01-16 - Enhanced Windows Setup Documentation

Added

📄 Improved Windows Setup Guide: Added specific instructions for installing and verifying make on Windows, streamlining the prerequisite setup for Windows users.

[v0.8.0] - 2025-01-15 - Documentation Enhancements and Version Update

Changed

📄 Improved Docker Setup Guide: Added troubleshooting information and a helpful resource link in the README.md to assist users with common Docker Desktop WSL update issues during setup.

[v0.7.0] - 2025-01-15 - Enhanced Local Development Experience

Added

✨ Introduced IntelliJ/PyCharm Run Configurations for LLMWrappingAgent, core API development, and core API testing, simplifying local setup and execution of key components.
🚀 Implemented a standardized start_service.sh script to streamline the local execution and setup of various microservices (e.g., agents, API), ensuring consistent dependency management and startup procedures.
📄 Created a dedicated .env file for the API development playground, simplifying local environment configuration and connection details for testing and development.

Refactor

🔄 Updated aihub_pipeline's dependency on aihub_lib to use a local path reference instead of a Git tag, facilitating monorepo development and simplifying inter-project dependency management.

Removed

🗑️ Dropped the bson dependency from aihub_lib, streamlining the library's overall dependencies.

Added

🦾 New LLM Wrapping Agent: Introduced LLMWrappingAgent, a basic agent designed to simplify the integration and usage of Large Language Models within agent workflows, including cost reporting capabilities.
📄 Comprehensive Testing Documentation: Added a detailed "Testing" section to the README.md, outlining the use of pytest and pytest-bdd for behavior-driven development (BDD) and structured testing of agents.

Changed

🔄 Restructured Playground Examples: Reorganized existing agent examples into a new minimal_workflow directory structure (e.g., conditional_workflow, context_workflow, simple_workflow) to improve clarity, consistency, and ease of understanding for different agent patterns.
📝 Enhanced Agent Documentation: Significantly revised and expanded Chapter 5 of the documentation (5_agents_in_detail.md) for improved readability and deeper explanations of agent concepts, context handling, and multi-agent coordination.
🧹 API Playground Reorganization: Consolidated and reorganized aihub_api/playground examples into clearer development and testing directories, streamlining API testing setups.

Refactor

⚡️ Unified LLM Agent Example: The previous DevAgent playground example has been refactored and integrated into the new LLMWrappingAgent within the basic agents directory, providing a standardized LLM interaction pattern.
⬆️ Improved Gitignore: Updated .gitignore to include common Python build artifacts, enhancing repository cleanliness.

Removed

🗑️ Deprecated Playground Agents and APIs: Removed several older, individual playground agent implementations and their corresponding API examples, streamlining the codebase by eliminating redundant or outdated examples.

[v0.4.0] - 2025-01-07 - Improved Onboarding and Automated Code Analysis

Added

✨ Introduced Comprehensive Code Quality Scans: Integrated SonarCloud into the CI/CD pipeline for the Agent, API, Lib, Pipeline, and Web core components, enabling automated code quality and security analysis.
🚀 New CI/CD Workflow for Analysis and Testing: Established a dedicated GitHub Actions workflow to automatically run SonarCloud scans on pull requests and pushes to main, enhancing continuous quality assurance and feedback.

Changed

📄 Expanded Developer Documentation: Significantly updated the main README.md with a detailed table of contents, extensive sections on project structure, repository types, branching strategy, and a comprehensive "Getting Started" guide, greatly improving developer onboarding and project clarity.

[v0.2.0] - 2025-01-07 - Enhanced Context Understanding and Pipeline Alignment

Changed

📄 Documentation Clarity: Clarified the behavior of RunContext during Human-in-the-Loop (HITL) pauses, emphasizing that it remains open to preserve agent state.
🔄 Context Utilization: Updated documentation to reflect that both RunContext and ThreadContext can be leveraged for maintaining agent state during HITL steps.

Refactor

🧹 Dependency Management: Transitioned aihub_pipeline's dependency on aihub_lib from a local path to a direct Git repository reference, streamlining module integration and future releases.

Monitoring & Alerting

Identity Provider Setup

Microsoft Entra ID

Sources

Changelog ​

[v0.292.2] - 2026-06-09 - Infrastructure Network Enhancements ​

Changed ​

[v0.292.0] - 2026-06-08 - Streamlined Python Distribution: PyPI Launch and Enhanced Documentation ​

Added ​

Changed ​

Refactor ​

[v0.291.13] - 2026-06-08 - Enhanced OpenAI Service Iframe Permissions ​

Changed ​

[v0.291.12] - 2026-06-08 - Refined Skill Permissions for Enhanced Control ​

Changed ​

[v0.291.11] - 2026-06-08 - Enhanced Changelog Documentation ​

Changed ​

[v0.291.10] - 2026-06-08 - Enhanced Security and Code Quality ​

Added ​

[v0.291.9] - 2026-06-08 - Simplified Access Control and Enhanced Identity Provider Setup ​

Added ​

Changed ​

Removed ​

[v0.291.8] - 2026-06-08 - Streamlined Initial Navigation and Enhanced Authentication ​

Added ​

Changed ​

Refactor ​

[v0.291.7] - 2026-06-03 - Enhanced Model Cost Reporting and API Robustness ​

Added ​

Changed ​

[v0.291.6] - 2026-06-03 - Core System Refinements and Performance Enhancements ​

Changed ​

Refactor ​

[v0.291.5] - 2026-06-02 - Enhanced FormKit Data Handling for Stability ​

Changed ​

Fixed ​

[v0.291.4] - 2026-06-02 - Enhanced Prompt Reliability for Image Rendering ​

Fixed ​

[v0.291.3] - 2026-06-02 - Stricter OAuth Role Defaults ​

Changed ​

[v0.291.2] - 2026-06-02 - GitHub Actions Reliability Improvement ​

Fixed ​

[v0.291.1] - 2026-06-02 - Improved Code Quality Checks with SonarCloud Action v6 ​

Changed ​

[v0.291.0] - 2026-06-02 - 🚀 Web Package Stability, Deployment, and Publishing Overhaul ​

Added ​

Changed ​

[v0.290.13] - 2026-06-02 - OAuth Default Role Access Update ​

Added ​

[v0.290.12] - 2026-06-02 - Documentation Refinements and Link Consistency ​

Changed ​

Refactor ​

[v0.290.11] - 2026-06-01 - Enhanced Documentation and Licensing Clarity ​

Added ​

Changed ​

[v0.290.10] - 2026-06-01 - Enhancing Project Governance: CLA and Contributor Guidelines ​

Added ​

Changed ​

[v0.290.9] - 2026-06-01 - Next-Gen Agents and Operational Excellence ​

Added ​

Changed ​

Removed ​

[v0.290.8] - 2026-06-01 - Introducing a Mixed-License Model and Enhanced Contribution Guidelines ​

Added ​

Changed ​

[v0.290.7] - 2026-05-29 - Introducing API Token Management and Documentation ​

Added ​

[v0.290.6] - 2026-05-29 - Backup Service Refinements and Code Quality Enhancements ​

Changed ​

Refactor ​

[v0.290.5] - 2026-05-28 - Dependency Security Enhancement ​

Security ​

[v0.290.4] - 2026-05-27 - Development Environment Update ​

Changed ​

[v0.290.3] - 2026-05-27 - Platform Refinements and Developer Experience Improvements ​

Added ​

Changed ​

Refactor ​

[v0.290.2] - 2026-05-27 - Introducing Comprehensive User Role Management ​

Added ​

Changelog

[v0.292.2] - 2026-06-09 - Infrastructure Network Enhancements

Changed

[v0.292.0] - 2026-06-08 - Streamlined Python Distribution: PyPI Launch and Enhanced Documentation

Added

Changed

Refactor

[v0.291.13] - 2026-06-08 - Enhanced OpenAI Service Iframe Permissions

Changed

[v0.291.12] - 2026-06-08 - Refined Skill Permissions for Enhanced Control

Changed

[v0.291.11] - 2026-06-08 - Enhanced Changelog Documentation

Changed

[v0.291.10] - 2026-06-08 - Enhanced Security and Code Quality

Added

[v0.291.9] - 2026-06-08 - Simplified Access Control and Enhanced Identity Provider Setup

Added

Changed

Removed

[v0.291.8] - 2026-06-08 - Streamlined Initial Navigation and Enhanced Authentication

Added

Changed

Refactor

[v0.291.7] - 2026-06-03 - Enhanced Model Cost Reporting and API Robustness

Added

Changed

[v0.291.6] - 2026-06-03 - Core System Refinements and Performance Enhancements

Changed

Refactor

[v0.291.5] - 2026-06-02 - Enhanced FormKit Data Handling for Stability

Changed

Fixed

[v0.291.4] - 2026-06-02 - Enhanced Prompt Reliability for Image Rendering

Fixed

[v0.291.3] - 2026-06-02 - Stricter OAuth Role Defaults

Changed

[v0.291.2] - 2026-06-02 - GitHub Actions Reliability Improvement

Fixed

[v0.291.1] - 2026-06-02 - Improved Code Quality Checks with SonarCloud Action v6

Changed

[v0.291.0] - 2026-06-02 - 🚀 Web Package Stability, Deployment, and Publishing Overhaul

Added

Changed

[v0.290.13] - 2026-06-02 - OAuth Default Role Access Update

Added

[v0.290.12] - 2026-06-02 - Documentation Refinements and Link Consistency

Changed

Refactor

[v0.290.11] - 2026-06-01 - Enhanced Documentation and Licensing Clarity

Added

Changed

[v0.290.10] - 2026-06-01 - Enhancing Project Governance: CLA and Contributor Guidelines

Added

Changed

[v0.290.9] - 2026-06-01 - Next-Gen Agents and Operational Excellence

Added

Changed

Removed

[v0.290.8] - 2026-06-01 - Introducing a Mixed-License Model and Enhanced Contribution Guidelines

Added

Changed

[v0.290.7] - 2026-05-29 - Introducing API Token Management and Documentation

Added

[v0.290.6] - 2026-05-29 - Backup Service Refinements and Code Quality Enhancements

Changed

Refactor

[v0.290.5] - 2026-05-28 - Dependency Security Enhancement

Security

[v0.290.4] - 2026-05-27 - Development Environment Update

Changed

[v0.290.3] - 2026-05-27 - Platform Refinements and Developer Experience Improvements

Added

Changed

Refactor

[v0.290.2] - 2026-05-27 - Introducing Comprehensive User Role Management

Added

Changed

Refactor

Security

[v0.290.1] - 2026-05-27 - Enhanced Dependency Stability and Configuration Flexibility