Highlights
- Collection Editor: direct CRUD on published chunks — create, update, delete, split, merge, reorder, and quality scoring.
- Collection Agent: AI-powered assistant for collection management with streaming SSE chat, tool use, and automated collection cleaning.
- File Upload Ingestion: ingest PDF, DOCX, and TXT files via multipart upload.
- Configurable Chunking: choose between LLM-powered semantic chunking and character-based splitting.
- Document Lineage: browse documents grouped by source with chunk counts and quality aggregation.
- Chat Playground: RAG-powered Q&A with citations and conversation history per collection.
- Agent Settings: configure LLM model and API key for the collection agent.
- Collection Cleaning: automated detection and removal of low-quality or dirty chunks.
Verify checklist
- Collection editor endpoints create, update, split, merge, and reorder chunks.
- Agent chat streams responses with tool calls via SSE.
- File upload creates a session from PDF/DOCX/TXT content.
- Chat endpoint returns answers with citations.