Skip to content

CHANGELOG.md

Latest commit

 

History

History
202 lines (141 loc) · 9.4 KB

CHANGELOG.md

File metadata and controls

202 lines (141 loc) · 9.4 KB

Changelog

All notable changes to the Odoo LLM Integration project will be documented in this file.

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

[Unreleased]

Fixed

  • Process with AI Button Reliability: Replaced unreliable bus notification with client action pattern (2025-12-02)

    • Bus/WebSocket notifications failed on cloud deployments with code 1006 connection issues
    • New approach uses ir.actions.client to navigate and open AI chat reliably
    • Added pendingOpenInChatter state to llm.store service for cross-navigation state
    • Removed redundant bus subscription code from chatter patch
    • Affected modules: llm_assistant (18.0.1.5.3), llm_thread (18.0.1.4.3)

  • Module Dependency Issue: Fixed prompt_id field being referenced in llm_thread module without dependency on llm_assistant (2025-11-26)

    • Moved prompt_id serialization from llm_thread/models/llm_thread.py to llm_assistant/models/llm_thread.py
    • llm_thread can now be installed standalone without llm_assistant
    • prompt_id handling in _thread_to_store() now properly resides in the module that defines the field

Added

  • Tool Event System: Real-time tool execution tracking with streaming events (2025-01-12)
    • Added tool_called event when tool execution begins (llm_tool/models/mail_message.py:120-128)
    • Added tool_succeeded event when tool completes successfully (llm_tool/models/mail_message.py:147-157)
    • Added tool_failed event when tool execution fails (llm_tool/models/mail_message.py:166-176)
    • Events include comprehensive tool data: tool_call_id, tool_name, arguments, status, result/error
    • Events are yielded through existing generator chain for automatic propagation to Fleek platform

Changed

  • Enhanced Tool Execution Flow: mail.message.execute_tool_call() now emits real-time events (2025-01-12)
    • Tool execution status is now broadcasted in real-time for UI updates
    • Maintains backward compatibility with existing message-based status tracking
    • Events automatically flow through yield from chain to Fleek broadcasting system

Technical Details

  • Tool events are generated at key execution points in execute_tool_call() method
  • Event structure follows consistent pattern with type and tool_data fields
  • Integration with Fleek platform enables WebSocket broadcasting to clients
  • Supports real-time tracking of image generation and other AI tool operations

[16.0-pr] - 2025-01-04 - Major Architecture Consolidation

Added

  • PostgreSQL Advisory Locking: Prevents race conditions in concurrent generation scenarios
  • Unified Generation API: New generate() method provides consistent interface for text, image, and other content types
  • Enhanced Message System: Added body_json field to mail.message for structured data storage
  • Indexed Role Field: New llm_role field for 10x faster message queries
  • Auto-Detection System: Automatic prompt argument detection and schema synchronization
  • Schema Source Transparency: Clear indication of schema sources in UI (Prompt vs Model vs None)
  • Loading State Management: Proper async handling and loading indicators in forms
  • Comprehensive Test Coverage: Tests for prompt arguments, thread schemas, and race condition fixes

Changed

  • Module Consolidation: Merged llm_resource functionality into llm_knowledge module
  • Prompt Integration: Moved llm_prompt functionality into llm_assistant module
  • Message Subtypes: Moved from separate module into base llm module
  • Tool Message Format: All tool data now stored in body_json instead of separate fields
  • Assistant Management: Enhanced with integrated prompt templates and testing capabilities
  • Generation Forms: Improved with automatic schema detection and better error handling

Removed

  • llm_resource module: Consolidated into llm_knowledge
  • llm_prompt module: Integrated into llm_assistant
  • llm_mail_message_subtypes module: Moved to base llm module
  • Deprecated API methods: Replaced with unified generation interface

Fixed

  • Race Conditions: Fixed async loading issues in media form components
  • Schema Computation: Eliminated inconsistencies between template and arguments
  • Form Loading: Prevented empty forms and incorrect field rendering
  • Context Management: Improved handling of context changes and reloads
  • Tool Execution: Better error handling and structured data storage
  • Performance Issues: Optimized database queries with indexed fields

Security

  • Enhanced Tool Consent: Improved security framework for tool execution
  • Role-Based Access: Strengthened permission-based tool access control

Migration Notes

  • Automatic Migration: All existing installations will be automatically migrated
  • Data Preservation: No loss of existing messages and tool execution history
  • Backward Compatibility: Maintains support for existing workflows during transition
  • Module Updates: Dependencies automatically updated to reflect consolidations

[Previous Releases]

Module Version History

Core Modules

  • llm: 16.0.1.3.0 (2025-01-04) - Message subtypes integration and role optimization
  • llm_assistant: 16.0.1.4.0 (2025-01-04) - Integrated prompt templates and enhanced testing
  • llm_thread: 16.0.1.3.0 (2025-01-04) - Role field optimization and PostgreSQL locking
  • llm_tool: 16.0.3.0.0 (2025-01-04) - Body_json refactoring and enhanced execution
  • llm_generate: 16.0.2.0.0 (2025-01-04) - Unified generation API and clean integration
  • llm_store: 16.0.1.0.0 (2025-01-02) - Vector store abstraction framework

Provider Modules

  • llm_openai: 16.0.1.1.3 (2025-01-04) - Enhanced tool support and API improvements
  • llm_anthropic: 16.0.1.1.0 (2025-03-06) - Anthropic provider enhancements
  • llm_ollama: 16.0.1.1.0 (2025-03-06) - Chat method parameter updates
  • llm_mistral: 16.0.1.0.0 (2025-01-02) - Mistral AI integration
  • llm_litellm: 16.0.1.1.0 (2025-03-06) - LiteLLM integration updates
  • llm_replicate: 16.0.1.1.0 (2025-03-06) - Replicate provider improvements
  • llm_fal_ai: 16.0.2.0.0 (2025-01-04) - Unified generate endpoint and schema storage

Knowledge & Vector Store Modules

  • llm_knowledge: 16.0.1.1.0 (2025-01-04) - Consolidated resource management and RAG
  • llm_chroma: 16.0.1.0.0 (2025-01-02) - ChromaDB vector store integration
  • llm_pgvector: 16.0.1.0.0 (2025-01-02) - PostgreSQL vector extension
  • llm_qdrant: 16.0.1.0.0 (2025-01-02) - Qdrant vector database integration

Specialized Modules

  • llm_mcp: 16.0.1.0.0 (2025-01-02) - Model Context Protocol support
  • llm_training: 16.0.1.0.0 (2025-01-02) - Fine-tuning capabilities
  • llm_tool_knowledge: 16.0.1.0.0 (2025-01-02) - Knowledge base tool integration

Historical Changes (2025)

January 2025

  • 16.0.1.2.0 (llm_thread) - LLM base module message subtypes integration
  • 16.0.1.0.1 (llm_tool) - Minor fixes and improvements

March 2025

  • 16.0.1.1.0 (llm_thread) - Tool integration in chat interface
  • 16.0.1.1.0 (multiple providers) - Chat method parameter updates

April 2025

  • 16.0.1.1.1 (llm_thread) - Method name consistency updates
  • 16.0.1.0.1 (llm_tool) - Additional fixes and improvements

Performance Improvements

Database Optimization

  • 10x Query Performance: Indexed llm_role field eliminates expensive subtype lookups
  • Reduced Complexity: Consolidated modules reduce maintenance overhead
  • Optimized Frontend: Direct field access instead of computed role checking

User Experience

  • Smoother Loading: Proper async handling prevents UI flashing
  • Real-time Updates: Enhanced streaming generation with live feedback
  • Better Error Handling: Comprehensive error messages and fallback handling

Migration Guide

From llm_resource to llm_knowledge

  1. Automatic Process: Migration script handles module transition
  2. Data Preservation: All resources, collections, and embeddings preserved
  3. API Compatibility: All existing methods continue to work
  4. Dependency Updates: Module dependencies automatically updated

From Separate Prompt Module

  1. Seamless Integration: Prompts now managed within assistants
  2. Enhanced Features: Auto-detection and testing capabilities added
  3. Template Compatibility: All existing templates continue to work
  4. Improved UI: Integrated prompt management in assistant interface

Tool System Migration

  1. Structured Data: Tool results now in body_json format
  2. Enhanced Execution: Better error handling and result storage
  3. MCP Compatibility: Improved Model Context Protocol integration
  4. Provider Support: Unified tool calling across all providers

Compatibility Matrix

Module Version Odoo Version Python Version Dependencies
16.0.x.x.x 16.0+ 3.8+ mail, web

Support & Resources

For more detailed technical information, see OVERVIEW.md for architecture details.