git-llm-utils

git-llm-utils enables automatically generating meaningful commit messages based on your staged changes using a LLM.

What it does

• Allows generating git commit messages from the repository staged changes.
• The commnit message can be generated with:
  • the standard git message workflow with a Git hook to capture your staged changes diff (the hook can be installed with a command).
  • with a commit command: git-llm-utils commit
• The diff message is generated with the configured llm (groq, ollama or any Litellm supported model).
• All configurations are stored using git config storage support.
• Helps you maintain clean, consistent, human-readable commit history without manual effort.

Why use it

• 🕒 Save time — no more writing commit messages for every change.
• ✅ Consistency — a unified style and format across commits.
• 📚 Clarity — automated summaries help communicate what changed and why.
• 🤝 Lower friction for contributors — beneficial in collaborative or open-source projects, or when onboarding new team members.

Getting started

Building from the source:

1. Clone the repo:

git clone [email protected]:pilardi/git-llm-utils.git

2. Build the environment (you'll need uv package manager installed in your system)

cd git-llm-utils && make install

3. Activate the environment and build the binary in your system

source .venv/bin/activate && make dist/git-llm-utils

4. Ensure your environment (or config) provides credentials to access a suitable LLM (see Configuration)

git-llm-utils verify

5. Go to a repository where you want to use the tool and stage some changes

cd _your_repository_path_here_ && git-llm-utils status

6. Commit the changes with the generated message: (all actions need to be confirmed by the user unless you pass the --no-confirm option before the action)

git-llm-utils commit

Using the binary

You can download the binary version correspoding to your OS from the release assets or build it locally using:

make dist/git-llm-utils

You'd only need to add it your local PATH afterwards.

Usage

See:

git-llm-utils --help

Configuration:

See:

git-llm-utils --config

by default the application uses ollama/qwen3-coder:480b-cloud as the model, therefore you'd need to have ollama running in your machine or configure a host where ollama would be running (using api_url).

Settings

All configuration settings are made using git config settings:

• emojis: True will allow the llm to generate Emojis in the commit message.
• comments: True will generate the message with commented lines.
• model: ollama/qwen3-coder:480b-cloud use Litellm syntaxis.
• reasoning: medium use Litellm Reasoning.
• max_input_tokens: how many tokens to send to the model at most (262144).
• max_output_tokens: how many tokens to request form the model at most (32768).
• api_key: None if not given, will read env variables according to Litellm api provider settings (ie: OPENAI_API_KEY).
• api_url: None (will use Litellm defaults, for ollama it's http://127.0.0.1:11434).
• description-file: README.md will use this file for context to the llm as as tool (if allowed).
• tools: False will allow the LLM to access the description of your repository.
• manual: True if True will not generate the commit message for every git commit when the commit hook is installed, unless the env.GIT_LLM_ON=True is set. Manual mode is active by default as you might want to select which commits needs llm comments.

Use:

git-llm-utils set-config setting --value value

to change the setting, where setting is any of the available settings.

Models

By default the system uses qwen3-coder:480b-cloud, however you can try other models such as nemotron-3-nano:30b-cloud. ie:

ollama pull nemotron-3-nano:30b-cloud && git-llm-utils --model ollama/nemotron-3-nano:30b-cloud commit

Local models are also supported however the quality of the message will drasticall vary based on the size of it (even worse for non-thinking models), ie:

ollama pull deepseek-coder:6.7b && git-llm-utils --model ollama/deepseek-coder:6.7b --reasoning none commit

References:

• Litellm is uses as the llm library to connect to your preferred LLM model server.
• Typer typer is the cli client library used to wire the application.
• Ollama is the default model server, however any litellm model api would work such as openai, groq or gemini.
• Git is the underlining supported repository, you'd need to have git installed in your system to use this tool.

---

Fun fact, all commit messages in this repo were generated by this tool

---

🤝 Contributing

Contributions are welcome! You can help by:

• Improving documentation
• Adding new hooks
• Improving prompt quality
• Reporting bugs or edge cases

---

📄 License

MIT License. See LICENSE for details.

---

✨ Author

Created by Pablo Ilardi GitHub: https://github.com/pilardi