sougur/graphrag repository overview

⁠CodeGraphAI: chat with your Python codebase

Upload a Python project as a ZIP and ask questions about it in plain English, such as "What calls place_order?" or "What could break if I change apply_discount?". The code is parsed (never executed) into a Neo4j knowledge graph of calls, inheritance and imports, and into a ChromaDB vector index with Cohere embeddings. Answers come from Groq, with every claim citing a real path:line range.

Source code: https://github.com/SOUGUR/GraphRAG⁠

⁠Run

The API listens on port 8000 and needs a Neo4j 5 database plus Groq and Cohere API keys. The easiest way to run it is the docker-compose setup in the GitHub repository. To run the image on its own:

docker pull sougur/graphrag
docker run -p 8000:8000 --env-file .env -v graphrag_data:/data sougur/graphrag

The health check is at /api/v1/health. Secrets are never baked into the image: pass them at runtime.