Language

English 简体中文

Knowledge base (RAG)

Router: app/server/routers/kdb.py. All route prefixes in the main doc use /api/knowledge-base.

How this ties to other features

• Agents: set kdb_ids in AgentConfigDTO.availableKnowledgeBases (or the snake_case alias) so the runtime can attach the right retrieval to that agent. Exact enforcement is in the chat/agent wiring layer, not the KDB router alone.
• Tools: some UIs go through a tool that wraps POST .../retrieve instead of calling HTTP from the client—same backend, different surface for auth/auditing.

Suggested integration order

1. POST .../add and store kdb_id.
2. POST .../doc/add_by_files to ingest files; capture taskId.
3. Poll .../doc/task_process and/or .../doc/list until processing is healthy; use .../redo paths when you need a rerun.
4. POST .../retrieve to validate RAG end-to-end.
5. Add the kdb_id to agent config or your own orchestration.

Common pitfalls

• Delete vs clear: DELETE /api/knowledge-base/delete/{kdb_id} vs POST .../clear have different semantics.
• Task IDs here are ingestion jobs for that KDB, not the /tasks scheduler in HTTP_API_TASKS.md nor the async work in HTTP_API_AGENT.md.
• Re-read the main doc’s table of query parameters for doc/list and doc/task_process before you script filters.

Back to HTTP API Reference