commitbro 0.1.0

Generate Git commit messages using local or remote AI models.

🧠 CommitBro

CommitBro is a privacy-first, local-friendly Git commit message generator that uses modern LLMs like Mistral, Starling, or TinyLlama — all running locally via Ollama, or optionally via a hosted FastAPI backend.

---

✨ Features

• 🔐 100% privacy-first — no data leaves your machine
• 🤖 Generates commit messages from git diff --cached
• 🎯 Instruction-tuned LLMs: Mistral, TinyLlama, Starling, Gemma, etc.
• 🧠 Optional training from your own commit history
• 🧰 FastAPI backend included (for training and expansion)
• 🌍 Supports multi-language commit messages
• 🐳 Docker-ready for container lovers

---

📦 Installation

1. Clone and Install (Dev Mode)

git clone https://github.com/yourusername/commitbro.git
cd commitbro
pip install -e .

Requires Python 3.12+, Ollama, and git.

2. Run Setup

commitbro-setup

Choose:

• Your preferred language (e.g. English, Spanish, Hindi)
• Whether to train models (locally, remotely, or skip)
• Your preferred model (mistral, tinyllama, etc.)

---

🚀 Usage

1. Stage your changes:

   git add .
   

2. Generate commit message:

   commitbro
   

3. Accept, regenerate, or cancel:

   Suggested Commit Message:
   > Fix user login state handling on timeout
   
   ✅ Accept this? [y/n/r]: 
   

4. If accepted, CommitBro will automatically commit your changes.

---

⚙️ Configuration

CommitBro stores user settings in:

~/.commitgen/config.json

Sample config:

{
  "language": "English",
  "training": "none",
  "remote_training_url": "",
  "model": "mistral"
}

You can change the model, language, or training config anytime by rerunning:

commitbro-setup

---

🧠 Model Support (via Ollama)

Run any local model supported by Ollama:

Model	Size	Speed	Notes
tinyllama	~1.1B	⚡⚡⚡	Very fast, small commits
mistral	7B	⚡	General purpose, accurate
starling	7B	⚡	Optimized for dialogue
phi	2.7B	⚡⚡	Compact + reasoning
gemma:2b	2B	⚡⚡	Google’s latest compact model

To pull models:

ollama pull mistral
ollama run mistral

---

🛠️ Training

Extract commit history and send to remote training server:

commitbro-train

Only enabled if you selected remote or both training during setup.

You can use the included FastAPI server for training endpoints (see below).

---

🌐 FastAPI Backend (Optional)

A minimal FastAPI server is included for collecting commit training data.

Run it locally:

cd fastapi_server
uvicorn main:app --reload

Available Endpoint:

• POST /train — Accepts commit history for training

{
  "commits": [
    {
      "hash": "abc123",
      "message": "Fix bug in login flow",
      "body": "Adjusted redirect logic in session manager"
    }
  ]
}

---

🐳 Docker Support

Run CommitBro inside a container (CLI mode):

docker build -t commitbro .
docker run --rm -v $(pwd):/app commitbro

Or to expose FastAPI:

docker run -p 8000:8000 commitbro uvicorn fastapi_server.main:app --host 0.0.0.0 --port 8000

---

📁 Project Structure

commitbro/
  ├── cli.py              # Main CLI logic
  ├── core.py             # Git diff + model API
  ├── train.py            # Training script
  ├── setup_config.py     # First-time setup wizard

fastapi_server/
  └── main.py             # Remote training API

pyproject.toml            # Modern packaging config
Dockerfile                # Optional container setup
README.md                 # You're reading it!

---

🧰 Packaging & Build

Uses hatchling and pyproject.toml (PEP 621).

Install locally:

pip install -e .

---

🧪 Roadmap

• CLI with diff-based message generation
• Language + training setup wizard
• Remote FastAPI training backend
• VS Code extension
• Frontend (Next.js + ShadCN UI)
• Auth + analytics-free SaaS version

---

🧑‍💻 Author

Built with 💚 by @aidataguy

---

📜 License

MIT License — do whatever you want, just keep it open.