ai-pdf-chatbot-langchain
Template app that lets you chat with PDF documents using LangChain, LangGraph, OpenAI and Supabase.
- Stars
- 16.6k
- Forks
- 3.2k
- Licence
- MIT
- Language
- TypeScript
- Last update
- 7 months ago
- Type
- App
This repository is archived and no longer maintained.
What it is
This is a starter project that lets you upload PDF files and ask questions about them in a chat window. It reads the PDFs, stores searchable vector embeddings in a Supabase database, and uses an OpenAI model (or another supported provider) to answer questions with references to the retrieved text. It is built as a template you can copy and adapt, and the README says it also accompanies the book Learning LangChain.
- Upload a PDF and ask questions about its contents in a chat interface
- Study how an ingestion graph and a retrieval graph are structured in LangGraph
- Swap or customize the agent's behavior by editing the backend graph files
- Inspect each step of the graph using LangGraph Studio during development
Great for
- Developers learning how LangGraph ingestion and retrieval graphs fit together
- Developers who want a working Next.js and TypeScript starting point for document Q&A
- Students following the Learning LangChain book
- Teams prototyping a PDF question-answering feature with their own OpenAI and Supabase accounts
Not ideal for
- Anyone wanting a ready-made product without coding or setup, since it needs API keys and a configured Supabase database
- Production use, because the README states the project is no longer actively maintained and issues and pull requests may go unanswered
- Users who need a non-OpenAI setup without checking LangChain's provider support themselves, since the README only says other providers may work
How to get it
- 1
Clone and build from source
You'll need: Node.js v18+ (v20 recommended) · Yarn (npm also mentioned, but the monorepo is set up for Yarn) · Supabase project with a documents table and match_documents function · OpenAI API key
Clone the repository
git clone https://github.com/mayooear/ai-pdf-chatbot-langchain.gitInstall dependencies from the monorepo root
yarn installCopy the backend environment template
cp backend/.env.example backend/.envCopy the frontend environment template
cp frontend/.env.example frontend/.env - 2
Run it
- Fill in OPENAI_API_KEY, SUPABASE_URL and SUPABASE_SERVICE_ROLE_KEY in backend/.env
- Start the backend with yarn langgraph:dev from the backend folder (port 2024 by default)
- Start the frontend with yarn dev from the frontend folder
- Open http://localhost:3000, upload a small PDF, then ask a question in the chat
- 3
Get the most out of it
- Enable LangSmith tracing with LANGCHAIN_TRACING_V2 and a LangSmith API key to debug each step
- Use LangGraph Studio to visualize and test the graph before changing the UI
- Adjust the agent logic in src/retrieval_graph.ts and src/ingestion_graph.ts
- Review src/shared/configuration.ts to change model providers or vector stores
Downloads
Strengths
- Clear separation of ingestion and retrieval logic into named graph files
- Includes a web UI with streaming responses, so you can see answers appear in real time
- MIT licensed, so you can copy and modify it freely
Trade-offs
- Archived and not actively maintained, so bugs and dependency updates are unlikely to be addressed upstream
- Requires several paid or account-based services (OpenAI, Supabase) and manual database setup
- The README is truncated in places, so some setup details must be found in linked documentation
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| ai-pdf-chatbot-langchain (this) | 16.6k | App | intermediate | 7 months ago | Developers learning how LangGraph ingestion and retrieval graphs fit together |
| anything-llm | 67.8k | App | beginner | 7 days ago | People who want a private ChatGPT-style assistant over their own PDFs and text files |
| langchain | 149k | Framework | intermediate | 4 days ago | Python developers building LLM-powered applications |
| langgraph | 44.3k | Library | intermediate | 4 days ago | Python developers building multi-step AI agents that must survive failures or long waits |
| botpress | 15k | Dev tool | intermediate | 7 days ago | Developers who want to write integrations and bots as TypeScript code using the Botpress CLI and SDK |
| markitdown | 193k | Library | beginner | 9 days ago | Developers building LLM or text analysis pipelines that need document text |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.