Skip to content

API chat endpoints

Laurian Gridinoc requested to merge api-updates into main

Upgraded from ancient langchain, added langserve to serve langchain chains (creates several endpoints where you can invoke or stream results, including a test playground).

  • FastAPI + langserve API key secured endpoints (/chat, /rag_chat) along "Proposal Insights" 1st step /similarity_search
  • moved from ChromaDB to PGVector
  • added PostgresChatMessageHistory

TODO where UX discussion is needed:

  • extend PostgresChatMessageHistory to have extra columns in the db table: for user id, other UI grouping/namespace, and other metadata (user feedback, TruLens/RAGAS evaluation of response)
  • add chat history retrieval endpoints, per user, per collection?/widget?/namespace?
  • add chat history delete/soft-delete/archive endpoint?
  • add LLM option in chats (now it defaults to OpenAI, which can be instrumented via a LLM Proxy to use a LocalAI endpoint - see LLM-infra)


  • control context window size (depends on LLMs used/selected by user?) with context compression
  • "duplicate" /rag_chat to "Proposal Insights"-like 2nd step document handling (map/reduce or rolling context compression)


  • add Langfuse instrumentation (to log queries, prompts, and even later control/swap prompts)

Merge request reports