GitAICommit

Rust-based CLI tool that generates semantically relevant commit messages using local AI models. Built with Rust, Tokio, and utilizing Ollama, it offers a privacy-focused, offline, and fast intelligent git repository analysis and AI-powered commit message generation.

Why GitAICommit?

Writing meaningful commit messages is tedious, but crucial for maintaining a healthy codebase. GitAICommit makes it easy by:

• Generating semantic commit messages automatically
• Understanding the context of your code changes
• Supporting conventional commit standards
• Using local AI models for private offline inference
• Providing a blazingly fast, Rust-based CLI interface
• Working across all major operating systems
• Offering flexible configuration options

Quick Start

Prerequisites

• Rust 1.70 or later
• Git
• Ollama (for local AI models)

Installation

# Install from source
git clone https://github.com/npiesco/gitAICommit.git
cd gitAICommit
cargo install --path .

Basic Usage

# Generate and commit changes with AI
git-ai-commit

# Stage all changes and generate commit message
git-ai-commit -a

# Preview commit message without committing
git-ai-commit -d

# Show verbose output
git-ai-commit -v

Demo

Command Line Options

USAGE:
    git-ai-commit [OPTIONS]

OPTIONS:
    -m, --model <MODEL>             AI model to use [default: gemma3:4b]
    -f, --max-files <COUNT>         Max files to include in diff [default: 10]
    -l, --max-diff-lines <LINES>    Max diff lines per file [default: 50]
    -d, --dry-run                   Show what would be committed
    -v, --verbose                   Show verbose output
    -p, --port <PORT>               Port for Ollama server [default: 11434]
    -t, --timeout <SECONDS>         Timeout for AI generation [default: 60]
    -a, --add-unstaged              Automatically stage all unstaged changes
        --confirm                   Ask for confirmation before committing
        --template <FILE>           Path to a custom prompt template file
        --list-models               List available Ollama models and exit

Configuration

Config File

Create ~/.config/git-ai-commit/config.toml:

model = "gemma3:4b"
max_files = 15
max_diff_lines = 100
timeout_seconds = 120
confirm = true

Environment Variables

# Override default model
export GIT_AI_COMMIT_MODEL=gemma3:4b

# Set custom Ollama port
export OLLAMA_HOST=http://localhost:11434

Features

AI-Powered Commit Messages

• Uses local Ollama models for privacy and offline use
• Generates semantic, conventional commit messages
• Understands code context and changes
• Supports multiple AI models
• Automatically downloads missing models

Fast & Efficient

• Built with Rust for maximum performance
• Asynchronous I/O with Tokio
• Minimal dependencies
• Low memory footprint

Smart Git Integration

• Analyzes staged and unstaged changes
• Understands file types and changes
• Supports partial staging
• Works with existing git workflows

Advanced Usage

Auto-Staging Behavior

When using the --add-unstaged flag, the tool will:

1. Stage all unstaged changes (including untracked files)
2. Generate a commit message based on all staged changes
3. Only proceed with the commit if there are actual changes to commit

Model Management

The tool will automatically download the specified model if it's not already available. Supported models include:

• gemma3:4b (default) - Best balance of performance and quality
• gemma3:1b - Faster but less accurate than 4b
• tinyllama:latest - Very fast but less accurate

To list available models:

git-ai-commit --list-models

Development

Running Tests

# Run all tests
cargo test

# Run specific test file
cargo test --test git_info_test
cargo test --test cli_args_test

Building for Release

# Build with optimizations
cargo build --release

License

MIT - See LICENSE for details.

Acknowledgments

• Ollama for local AI models
• git2-rs for Git bindings
• Clap for CLI parsing
• Tokio for async runtime