git-auto-commit

Generate Git commit messages from staged changes using a local Ollama model.

Features

• Reads staged changes and asks Ollama for a commit message
• Prompts for confirmation before running git commit
• Auto-selects an installed chat model when none is configured
• Configurable commit message format and Ollama model
• Skips common build artifacts, lockfiles, and binary files by default
• Summarizes very large diffs when the prompt would exceed 8,000 characters
• Detects bootstrap commits (many new files) and adjusts the prompt

Quick start

You need Git and Ollama. You do not need Rust or Cargo.

1. Install git-auto-commit

On macOS or Linux, open a terminal and run:

curl -fsSL https://raw.githubusercontent.com/usamakhangt4/git-auto-commit/main/install.sh | sh

On Windows, open PowerShell and run:

irm https://raw.githubusercontent.com/usamakhangt4/git-auto-commit/main/install.ps1 | iex

2. Set up Ollama

Start Ollama and download a model:

ollama pull llama3.2:3b

3. Use it in any Git project

git add .
git-auto-commit

The tool generates a commit message and asks for confirmation before committing.

Other installation options

Download manually

Download the archive for your operating system from GitHub Releases, extract it, and place the binary in a directory on your PATH.

To choose another installation directory, set GIT_AUTO_COMMIT_INSTALL_DIR before running one of the quick-install commands above.

Build from source

This option requires Rust 1.75 or newer.

git clone https://github.com/usamakhangt4/git-auto-commit.git
cd git-auto-commit
cargo build --release

Copy the binary to a directory on your PATH:

cp target/release/git-auto-commit ~/.local/bin/

Ensure ~/.local/bin is on your PATH.

Install with Cargo

If you already have Rust installed:

cargo install --path .

Usage

Stage changes, then run the tool inside a Git repository:

git add .
git-auto-commit

Example session:

Using model: llama3.2:3b | Format: type(scope): description

Generated Message:
>>> feat(cli): add commit message generator

Commit? [y/N]: y
Successfully committed!

Answer y or Y to commit; anything else aborts.

Subcommands

git-auto-commit set-model llama3.2:3b
git-auto-commit set-format "type(scope): description"
git-auto-commit --help
git-auto-commit --version

set-model and set-format work outside a Git repository.

Configuration

Config files live in ~/.config/git-auto-commit/ (or the platform equivalent):

File	Purpose
model.txt	Ollama model name (e.g. llama3.2:3b). Empty or missing → auto-discovery
format.txt	Commit message format instruction for the model
ignore.txt	Path patterns to exclude from the prompt (one per line; # comments allowed)

Precedence for format: format.txt → COMMIT_FORMAT env var → default type(scope): description.

Default ignore patterns include target/, node_modules/, lockfiles, minified assets, and common binary extensions. If ignore.txt is missing or empty, those defaults apply.

Environment variables

Variable	Default	Description
OLLAMA_HOST	http://localhost:11434	Ollama API base URL
COMMIT_FORMAT	type(scope): description	Format when format.txt is absent
GIT_AUTO_COMMIT_TIMEOUT_SECS	300	HTTP timeout for Ollama requests (seconds)
GIT_AUTO_COMMIT_CONFIG_DIR	(platform config dir)	Override config directory

Development

cargo test
cargo clippy --all-targets --all-features -- -D warnings
cargo build --release

License

MIT — see LICENSE.