RepoMind

A RAG-powered CLI that lets you ask natural language questions about any public GitHub repo — with file path citations in every answer.

$ repomind

  repomind v2.0  |  Anthropic  |  claude-sonnet-4-6
  Type /help for commands.

repo > https://github.com/expressjs/express
  Fetched 212 files
  Chunked into 548 chunks
  Embedding done

[expressjs/express] > How is routing implemented?

── ◆ thinking ────────────────────────────
The user is asking about routing internals...
──────────────────────────────────────────

Routing uses a **Layer** system. Each `Route` is an isolated middleware
stack defined in `lib/router/layer.js`. The `Router` matches paths and
delegates to `Route` instances in `lib/router/index.js`...

──────────────────────────────────────────

[expressjs/express] > exit

repo > /repo        ← switch to another indexed repo instantly
repo > exit
  1 collection deleted.

Demo

Coming soon

How it works

1. Fetch — pulls the repo file tree via GitHub Trees API (no git clone), fetches raw content in parallel
2. Chunk — splits files into ~400-token chunks with overlap
3. Embed — encodes chunks locally with all-MiniLM-L6-v2 and stores in ChromaDB
4. Chat — on each question, retrieves the top relevant chunks and sends them as context to the LLM
5. Cleanup — all ChromaDB collections deleted automatically when the session ends

Claude responses optionally use extended thinking — the reasoning process streams live in the terminal before the final answer. Toggle it anytime with /config.

Setup

1. Clone and create the environment

git clone https://github.com/moliu0104/RepoMind.git
cd RepoMind
conda create -n repomind python -y
conda activate repomind
pip install -r requirements.txt
pip install -e .

2. Run

repomind

On first launch, a setup wizard walks you through:

• Provider (Anthropic / OpenAI / MiniMax / Ollama)
• Model (arrow-key select)
• API key (hidden input)
• Extended thinking on/off (Anthropic only)

Config is saved to ~/.repomind/config.json. Run /setup anytime to change it.

Commands

| Command | Description | |---|---| | /help | List all commands | | /setup | Re-run the setup wizard | | /config | Toggle extended thinking and view current settings | | /repo | Arrow-key picker — switch between repos indexed this session | | /clear | Clear the terminal | | exit | Inside chat → back to repo >; at repo > → quit |

Docker

# Build (embedding model baked in — no network needed at runtime)
docker build -t repomind .

# Run with Claude
docker run -it -e ANTHROPIC_API_KEY=sk-ant-... repomind

# Run with OpenAI
docker run -it -e OPENAI_API_KEY=sk-proj-... repomind

Stack

| Component | Library | |---|---| | Embeddings | sentence-transformers — all-MiniLM-L6-v2 (local, free) | | Vector DB | ChromaDB (local, auto-deleted on exit) | | LLM | Anthropic Claude or OpenAI (configured via wizard) | | Terminal UI | rich — Markdown rendering, live thinking stream | | Menus | questionary — arrow-key selection throughout | | GitHub | Trees API + raw.githubusercontent.com |