Doc Oc
Talk to your codebase

• RAG enhanced LLM that utilizes an indexed Github repo as its content store.
• Used Cohere's embed-english-v3.0 embedding model to create embeddings for repository files, and for user query.
• Used Pinecone as vectorstore and Supabase to store actual file contents.
• Used Cohere's command-r as LLM that recieves context.

To run backend:

1. Use pip to install pipenv, the package manager for the backend: pip install pipenv
2. Run pipenv install to install dependencies
3. Get api keys for Cohere, Pinecone, Github, and Supabase (key and url) and put them into a .env file.
4. Run run.py script to start server. You can also type uvicorn main:app --reload --port 8000 in the terminal.

To run frontend:

1. Run npm install to install dependencies.
2. In the terminal, type npm run dev to run in dev mode or run npm run build and npm run start to run in production mode.

• Pinecone - Create a pinecone index called "doc-oc" with dimensions as 1024.

• Supabase - Create two tables:

  • github_repos - Stores all the index repos

    • Columns: id (PK, int8), repo_path (varchar)

  • files - Stores data related to files

    • Columns: id (PK, int8), repo_id (FK, int8), name (varchar), path (varchar), content (text)