Update training and service listing guidelines in chat and summarizer systems: Adjust the training year to 2025-2026, enhance the structure of service and training listings to include context-specific arguments, and clarify the use of contact information. Ensure that only verified details are provided and emphasize the importance of using catalog resources exclusively.

Add support for .docx file ingestion: Update requirements to include python-docx, modify document type detection to support .docx, and implement text extraction and indexing for Word documents in the ingestion service.

Update pricing settings for 'mistral-small-latest' and add new model 'mistral-small-2603': Adjust input and output rates for the 'mistral-small-latest' model and introduce pricing for the new 'mistral-small-2603' model to enhance cost management in the agent's operations.

Update agent model to 'mistral-small-2603': Change the model used in both the project and conversation subagents to 'mistral-small-2603' for improved performance and efficiency in summarization tasks.

Add middleware factories for LangChain agents: Introduce new middleware functions for main and project summarizer agents, enhancing the agent's capabilities with context editing and summarization features. Update retrieval tools and agent workflows to integrate these middlewares, ensuring improved message handling and processing. Add tests to validate middleware functionality and defaults.

Add sources parameter to completion requests and update documentation: Enhance the CompletionRequest model to include an optional sources field for document filtering. Update related API examples and documentation to reflect this change, ensuring clarity on usage and implications for project document retrieval.

Refactor message handling in AgentService: comment out text extraction for AIMessage to streamline processing, while maintaining document metadata extraction for AIMessage and ToolMessage types. This change aims to improve clarity and maintainability of the code.

Implement PDF summary generation tool: add a new tool for generating project-aware PDF summaries from conversations, enhancing the summarization capabilities of the agent. Update retrieval tools for improved metadata handling and streamline the agent's tool integration. Refactor PDF generation logic to handle link sanitization and improve error handling during PDF creation.

Refactor chat system prompt and tools: enhance resource organization by clarifying the structure of available resources, including services and training, and improve the 'check_project_id' tool to return None when no project is identified. Remove redundant session context from the agent's system prompt for cleaner implementation.

Update chat system prompt guidelines: remove references to similarity scores and 'slug', clarify service listing format by omitting unnecessary details, and ensure professional tone in responses.

Fix error replacement

Update dependencies and enhance agent functionality: upgrade LangChain to v1.2.x, adjust retrieval tools for improved pagination and filtering, and implement new project ID checks. Modify settings for RAG top-k and add support for custom state management in agents. Enhance documentation for streaming SSE client and tools policy, ensuring clarity on usage and constraints.

Enable V2 again

Update default agent identifier from 'V2' to 'V1' across multiple services for consistency in agent orchestration and handling.

Enhance chat tools agent: update tool call handling to prevent duplicate IDs by ensuring proper management of additional_kwargs, improving compatibility with Mistral API and overall message integrity.

Refactor chat tools agent: normalize tool call history and AI message handling to enhance conversation flow and maintain consistency in tool interactions.

Enhance chat tools agent: implement unique ID assignment for tool calls to ensure compatibility with Mistral API, improve handling of tool call history, and update fallback ID generation for tool calls. Refactor conversation message handling to maintain safe history.

Implement V2 workflow for agent orchestration, including new chat tools agent with retrieval capabilities, classifier node enhancements, and summarizer integration. Add detailed documentation for V2 features and prompts, ensuring improved agent interaction and context handling.

Refactoring gestion Agent

enhance transcription: add detailed logging for transcription job creation and processing, improve error handling for file pointer resets, and refine chunking logic for large audio files to ensure robust handling of transcription requests.

refactor transcription: replace Supabase quota checker with a noop implementation, update transcription models to include server-measured audio duration, and enhance job response structure for improved reliability and consistency.

feat transcription: implement meeting transcription quotas with project ID and duration limits, enhancing validation and quota management. Update settings and documentation to reflect new features and usage examples.

refactor transcription: update meeting transcription model to 'gpt-4o-mini-transcribe', enhance TranscriptionResponse and JobStatusResponse models with metadata fields, and implement usage normalization utilities for improved tracking and reporting.

fix settings: update meeting transcription model name to 'gpt-4o-transcribe' for consistency and improved functionality.

fix impact_service: update trace_llm_impact function to return value directly when min and max are not available, improving error handling and output consistency.

fix transcription: update meeting transcription model name to 'gpt-4o-transcribe-diarize' in settings and related files for consistency and improved functionality.

feat transcription: introduce asynchronous transcription job endpoints for audio and meeting files, allowing users to upload files and poll for job status. Update models and services to handle job creation, status tracking, and response formatting. Enhance documentation with examples and details on supported audio formats and job management.

feat transcription: update audio file upload limits to be configurable via settings, enhancing flexibility for transcription service. Update documentation to reflect new maximum file size and chunking behavior for large audio files.

chore: update Dockerfile to include ffmpeg dependency for enhanced audio processing capabilities

feat transcription: implement audio file chunking for transcription service to handle files exceeding 25 MB limit. Add pydub dependency for audio processing and enhance transcribe method to manage large audio files by splitting them into smaller segments. Update documentation for clarity on new functionality.

feat documents: add document deletion functionality and enhance upload process with UUID validation. Implement checks for duplicate documents during upload and update models to support delete responses. Refactor ingestion service to include document ID in processing pipelines.

feat documents: enhance document ingestion to support both PDF and plain text formats. Implement a new detection function for document types, update the upload endpoint description, and refactor the ingestion service to handle text files. Add tests for text file ingestion to ensure functionality.

feat transcription: add dedicated meeting transcription endpoint with support for audio file uploads and language options. Implement validation for file formats and sizes, and update settings for the new transcription model. Enhance API documentation and examples for clarity.

Enhance PDF generation and update configurations: Add PDF_LOGO_PATH to .env.example, include markdown-pdf in requirements.txt, and refactor PDF generation logic to support logo integration. Introduce tests for PDF generation from Markdown input and update summarizer agent to streamline export functionality.

feat voice: update documentation for `trigger_on_push` mode to include JSON message format for triggering flush. Simplify response handling in LangGraphProcessor by sending complete responses in a single block instead of segments, enhancing efficiency in transcript delivery.

Update .env.example and requirements.txt: Remove commented-out Twilio TURN configuration and add DAILY_API_KEY to the environment example. Update requirements to include daily-python for Linux compatibility.

feat voice: implement real-time response segmentation for agent replies, enhancing transcript delivery via WebRTC data channels. Update LangGraphProcessor to split responses into segments and modify frontend to display these segments in real-time. Update documentation to reflect new segment handling features.

feat voice: implement real-time segment handling in `trigger_on_push` mode, enhancing transcript delivery via WebRTC data channels and app messages. Update `TranscriptBufferProcessor` to emit transport messages for each buffered segment, and adjust documentation to reflect these changes.

fix voice: handle None response in TTS cleaning function and remove max length restriction for response cleaning

feat voice: integrate WebRTC ICE server configuration, enhance voice pipeline with new modes, and improve VAD settings for better audio processing

Refactor voice bot and enhance capabilities endpoint: Consolidate voice bot logic into a single function, update transport handling for Daily and SmallWebRTC, and introduce a new capabilities endpoint to check transport availability. Update documentation to reflect these changes.

Add minimal Daily.co voice test page: Introduce a new endpoint for a simplified UI without iframe, and update documentation to reflect the new minimal page and its usage alongside the existing Daily.co test page.

Enhance voice bot configuration: Set audio output sample rate to 24000 for compatibility with Daily TTS encoding. Remove audio recording functionality from the pipeline to streamline audio transport.

With daily

Update voice API configuration: Add Twilio TURN server support and enhance ICE server retrieval logic. Update environment example and settings to include Twilio credentials. Introduce comprehensive client integration documentation for voice API.

Update Dockerfile: Upgrade Python version from 3.12 to 3.13 for improved performance and features.

Implement WebRTC ICE server configuration: Add support for dynamic and static ICE server retrieval in voice services, update FastAPI app to initialize voice handler, and enhance voice route endpoints. Update frontend to fetch ICE servers for WebRTC connections. Add comprehensive client integration documentation for voice API.

Update Dockerfile: Install libgl1 for graphical support and clean up apt cache to reduce image size.

Implement Voice Activity Detection (VAD) configuration: Add VAD parameters to settings, create VADConfigDTO for validation, and integrate VAD handling in the voice bot. Update frontend to allow dynamic VAD parameter adjustments during sessions.

Merge branch 'voice-1' into main