commit-bot 1.0.1rc1

A brief description of your project.

Commit Pilot

Table of Contents

• Description
• Features
• Installation
• Usage
• Configuration
• License

Description

Commit Pilot is a command-line tool that uses Large Language Models (LLMs) to automatically generate git commit messages from your staged changes. It streamlines the commit process by providing well-formatted, context-aware messages, which you can then accept, edit, or regenerate on the fly.

Features

• AI-Powered Commit Messages: Automatically generates descriptive commit messages from your code diffs.
• Interactive Workflow: Allows you to accept, edit, regenerate, or change models before committing.
• Multiple Backends: Supports both Ollama and VLLM for running local LLMs.
• Configurable Models: Easily configure which model to use and its generation parameters.
• Automatic Server Management: Can automatically start and stop the vLLM server to manage resources.

Installation

Prerequisites

• Python 3.10
• Git
• Docker and Docker Compose (for the Ollama backend)

1. Backend Setup

You must have a model backend running. Choose one of the following:

Ollama Backend

1. Start the Server: Use Docker Compose to launch the Ollama server.

   docker-compose up -d
   

2. Pull Models: docker-compose will automatically execute pull_ollama_model.sh to download the default models.

VLLM Backend

For the VLLM backend, you must manually download model weights from Hugging Face. The application will automatically start and stop the VLLM server as needed.

• where to put the model weights:
  • You can put the model weights in ./src/commit_pilot/model_weights/
  • Or specify a different path in ~/.config/commit_pilot/job.conf under the key vllm_model_weights_root_dir.
    • You can create this config file by copying src/commit_pilot/conf/job.conf to ~/.config/commit_pilot/job.conf and modifying it as needed.

(For more details on backend setup, see the README in src/commit_pilot/conf/)

2. Install the Package

Clone the repository and use pip to install the package in editable mode.

git clone <repository-url>
cd commit_pilot
pip install -e .

Usage

Once installed, simply run the commit-pilot command in your git repository after staging your changes (git add .).

commit-pilot

This will start an interactive session where the tool generates a commit message and prompts you for action:

• (y)es: Accept the message and commit.
• (n)o: Abort the commit.
• (s)how: Show the staged diff and the generated message again.
• (r)egenerate: Generate a new commit message.
• (m)odel: Change the LLM used for generation.
• (e)dit: Manually edit the commit message in your default text editor.

Configuration

The behavior of Commit Pilot is controlled by two configuration files located in src/commit_pilot/conf/:

• job.conf: Defines runtime settings, such as the currently active model, server timeouts, and resource limits (e.g., GPU utilization for VLLM).
• model.conf: Acts as a catalog for all available models, defining their connection details, backend type (Ollama, VLLM, or third-party), and default generation parameters.

For detailed information, please refer to the Readme.md inside the src/commit_pilot/conf/ directory.

License

This project is licensed under the MIT License. See the LICENSE file for details.