Codebase RAG Assistant is a Streamlit-based application that lets you quickly index a GitHub repository, store the code as vector embeddings in Pinecone, and then retrieve relevant context to answer questions about the codebase using an LLM (GPT-4o). Perfect for large projects that need fast, intelligent codebase querying.
- Clone – User enters a GitHub repo URL. The app clones the repo locally, reads relevant files, and splits them into chunks.
- Embed & Store – Vector embeddings are created for each chunk and stored in Pinecone under a unique namespace.
- Wait for Ingestion – The app polls Pinecone to ensure the vector count matches the number of chunks added.
- Query – Once ingestion completes, users can ask questions via the chat. The top results from Pinecone are pulled into a GPT-4o prompt, which returns a context-based answer.
- Streamlit for the UI and user interactions.
- GitPython for cloning GitHub repositories locally.
- LangChain modules:
langchain_openaifor GPT-4o embeddings and chat completion,langchain.text_splitterfor chunking code into smaller pieces,langchain.docstore.documentto structure code chunks as documents.
- Pinecone for vector database storage and retrieval.
