// the find
mayooear/ai-pdf-chatbot-langchain
AI PDF chatbot agent built with LangChain & LangGraph
A monorepo template for PDF-ingestion chatbots using LangGraph state machines, Supabase for vector storage, and Next.js on the frontend. It's a companion project to the O'Reilly LangChain book, so it's tutorial-grade code, not production infrastructure. Explicitly unmaintained as of the README notice.
The two-graph separation (ingestion vs. retrieval) is architecturally clean — you can swap either graph independently. LangGraph's state machine approach makes the agent flow inspectable and debuggable via LangSmith tracing, which is genuinely useful when something goes wrong mid-pipeline. The frontend/backend split via Turborepo is set up correctly and the SSE streaming integration works end-to-end. There are actual tests, including an integration test for the retrieval graph, which is more than most tutorial projects bother with.
The repo is explicitly abandoned ('not actively maintained') — that's a hard stop for anything you'd run in production; LangGraph's API has been moving fast and breakage is likely. Chat history is ephemeral per session with no persistence, so any real use case requires rebuilding that layer from scratch. Hard dependency on Supabase as the only vector store makes swapping to Pinecone, pgvector, or anything else a non-trivial refactor despite the README promising it's configurable. The 5-PDF / 10MB cap is hardcoded in the ingest route with no server-side queue, so concurrent uploads will just race.