• LLM-Powered Commit Messages: Automatically generates clear, concise, and context-aware commit messages using large language models.
• Deep Contextual Analysis: Understands your code by analyzing staged changes, repository structure, and recent commit history to provide highly relevant suggestions.
• Multi-Provider & Model Support: Flexibly works with leading AI providers (Anthropic, Cerebras, Gemini, Groq, OpenAI, OpenRouter, Streamlake/Vanchin, Z.AI) and local providers (LM Studio, Ollama), easily configured through an interactive setup or environment variables.
• Seamless Git Workflow: Integrates smoothly into your existing Git routine as a simple drop-in replacement for git commit.
• Extensive Customization: Tailor commit messages to your needs with a rich set of flags, including one-liners (-o), AI hints (-h), scope inference (-s), and specific model selection (-m).
• Streamlined Workflow Commands: Boost your productivity with convenient options to stage all changes (-a), auto-confirm commits (-y), and push to your remote repository (-p) in a single step.
• Interactive Reroll with Feedback: Not satisfied with the generated commit message? Use r for a simple regeneration, or r <feedback> to provide specific improvement suggestions (e.g., r make it shorter, r focus on the bug fix).
• Token Usage Tracking: Display token consumption statistics (prompt, completion, and total tokens).
• Security Scanner: Built-in detection of secrets and API keys in staged changes to prevent accidental commits of sensitive information.

gac analyzes your staged changes to generate high-quality commit messages with the help of large language models. The tool uses a sophisticated prompt architecture that separates system instructions from user data, enabling better AI understanding and more consistent results.

• Dual-Prompt System: GAC uses a separated prompt architecture where system instructions (role definition, conventions, examples) are sent as system messages, while git data (diffs, status) are sent as user messages. This follows AI best practices for improved model performance.
• Smart Context Analysis: The tool examines your repository structure, recent commit history, and README files to understand the broader context of your changes.
• Intelligent Diff Processing: Large diffs are automatically preprocessed to focus on the most important changes while staying within token limits.

git add .
gac

You can try gac without installing it using uvx:

# Try gac without installation
uvx gac --help

# Set up configuration (creates ~/.gac.env)
uvx gac init

# Use gac on staged changes
git add .
uvx gac

Install system-wide using pipx from the GitHub repository:

# Install pipx if you don't have it
python3 -m pip install --user pipx
python3 -m pipx ensurepath

# Install gac
pipx install gac

Verify installation:

gac --version

The recommended way to configure gac is using the interactive setup:

gac init

This command will guide you through selecting an AI provider, model, and securely entering your API keys. It will create or update a user-level configuration file at $HOME/.gac.env.

Example $HOME/.gac.env output:

GAC_MODEL=anthropic:claude-3-5-haiku-latest
ANTHROPIC_API_KEY=your_anthropic_key_here

Alternatively, you can configure gac using environment variables or by manually creating/editing the configuration file.

You can manage settings in your $HOME/.gac.env file using gac config commands:

• Streamlake uses inference endpoint IDs instead of model names. When prompted, paste the exact endpoint ID from the Streamlake console.

• For local providers like Ollama and LM Studio, gac will ask for the base API URL. API keys are optional for these providers unless your instance requires authentication.

• Show config: gac config show

• Set a value: gac config set GAC_MODEL groq:meta-llama/llama-4-scout-17b-16e-instruct

• Get a value: gac config get GAC_MODEL

• Unset a value: gac config unset GAC_MODEL

Test that gac is working properly with your configuration:

# Make a change to a file
echo "# Test change" >> README.md
git add README.md
gac -o # Generate a one-line commit message

You should see an AI-generated commit message.

To upgrade gac to the latest version, run:

pipx upgrade gac

Once installed and configured, using gac is straightforward:

1. Stage your changes:

   git add .

2. Run gac:

   gac

   This will generate a commit message for review. Confirm with y to accept the message.

• Generate a commit message: gac
• Auto-accept the commit message: gac -y
• Stage all changes and generate a commit message: gac -a
• Generate a one-line commit message: gac -o
• Add a hint for the AI: gac -h "Fixed the authentication bug"
• Push the commit (requires accepting the commit message): gac -p
• Advanced usage: Add all, auto-confirm, push a one-liner with a hint: gac -aypo -h "update for release"
• Skip security scan: gac --skip-secret-scan (use with caution)

GAC includes a built-in security scanner to prevent accidental commits of sensitive information:

• Automatic Scanning: By default, scans all staged changes for potential secrets and API keys
• Interactive Protection: When secrets are detected, you can:
  • Abort the commit (recommended)
  • Continue anyway (not recommended)
  • Remove affected files and continue
• Wide Coverage: Detects AWS keys, GitHub tokens, OpenAI API keys, database URLs, private keys, and more
• Smart Filtering: Ignores example keys, placeholders, and test values to reduce false positives

To disable the security scan (not recommended unless you know what you're doing):

# Skip scan for one command
gac --skip-secret-scan

# Or disable via environment variable
export GAC_SKIP_SECRET_SCAN=true
gac

For a full list of CLI flags, advanced options, and example workflows, see USAGE.md.

• gac loads configuration from two locations (in order of precedence):
  1. User-level $HOME/.gac.env (applies to all projects for the user)
  2. Project-level .env (in the project root, overrides user config if present) Environment variables always take final precedence over both files.
• Keep API keys out of version control
• For troubleshooting, see docs/TROUBLESHOOTING.md

We welcome contributions! Please see docs/CONTRIBUTING.md for guidelines.

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

For questions, suggestions, or support, please open an issue or discussion on GitHub.