RAG API

Base path: /api/v1/projects/{project_id}/rag

Endpoints

Method	Endpoint	Description
POST	/projects/{id}/rag/query	Query knowledge base
POST	/projects/{id}/rag/index	Build/rebuild index
POST	/projects/{id}/rag/index/stream	Build index (SSE streaming progress)
GET	/projects/{id}/rag/stats	Index statistics
DELETE	/projects/{id}/rag/index	Delete index

Query Request

{
  "question": "What is attention mechanism?",
  "top_k": 10,
  "use_reranker": true,
  "include_sources": true
}

• question — The question to answer (required)
• top_k — Number of chunks to retrieve (default: 10)
• use_reranker — Apply reranker for relevance (default: true)
• include_sources — Include source chunks in response (default: true)

Query Response

{
  "answer": "LLM-generated answer with citations",
  "sources": [
    {"paper_id": 1, "chunk_id": "...", "score": 0.9}
  ],
  "confidence": 0.0
}

Index Stream (SSE)

POST /projects/{id}/rag/index/stream — Rebuild the vector index with Server-Sent Events for progress updates.

Response: text/event-stream

Event types:

Event	Description	data
progress	Indexing progress	{ stage, percent, message? }
complete	Indexing finished	{ indexed, collection, papers_updated }
error	Error occurred	{ message }

Delete Index

DELETE /projects/{id}/rag/index — Delete the vector index for the project. Returns ApiResponse[dict] with deletion result.