smart-commit

Install and use

Linux on x86_64. The installer does not need sudo; the setup asks before any step that does.

Install

$ curl -fsSL https://raw.githubusercontent.com/Lucas-BRT/smart-commit/main/install.sh | bash

This installs version 0.14.0. It lists your hardware, asks before anything that needs sudo, downloads the model and ends with a benchmark that picks the fastest device.

The default model is qwen2.5-coder-1.5b, 1.1 GB. Installing, updating and downloading models use the network; writing a message does not.

For a walkthrough, see the tutorial.

Drivers are installed with apt; Ubuntu 24.04 and derivatives are tested.

Requirements

  • Platform: Linux on x86_64, with glibc 2.35 or newer, and git.
  • Downloaded by the setup: the default model, 1.1 GB, and the pinned llama.cpp build, 31.6 MB.
  • Models and llama.cpp go to ~/.local/share/smart-commit; caches go to ~/.cache/smart-commit.

Use

git add -p
# print the message
smart-commit
# review it, then commit
smart-commit --commit

The message follows the style of your history.

On every commit

smart-commit hook install
# the editor opens with the message written
git commit

It works from any git client, and never blocks a commit: once the binary is gone, it does nothing.

More

Devices, models, settings, troubleshooting and uninstalling are in the usage guide.