autocommitmsg

Generates a commit message from a git diff using AI.

Features

• Supports OpenAI-like providers.
• Switch model based on the diff.

Installation

devenv users

Add the autocommitmsg input to the devenv.yaml file:

inputs:
  autocommitmsg:
    url: github:sestrella/autocommitmsg
    overlays: [default]
  nixpkgs:
    url: github:cachix/devenv-nixpkgs/rolling

Add the autocommitmsg hook to the devenv.nix file as follows:

{ pkgs, lib, ... }:

{
  dotenv.enable = true;

  git-hooks.hooks.autocommitmsg = {
    enable = true;
    entry = lib.getExe pkgs.autocommitmsg;
    stages = [ "prepare-commit-msg" ];
  };

  cachix.pull = [ "sestrella" ];
}

Note: Enabling dotenv is optional if the OPENAI_API_KEY environment variable is available.

Configuration

autocommitmsg can be configured via a .autocommitmsg.yaml file in the project's root directory or the user's home directory. The available configuration parameters are:

• base-url: The base URL of the OpenAI-like provider.
  • Default: https://generativelanguage.googleapis.com/v1beta/openai
• api-key: The name of the environment variable that contains the API key.
  • Default: GEMINI_API_KEY
• short-model: The model to use for diffs with fewer lines than diff-threshold.
  • Default: gemini-2.5-flash-lite
• long-model: The model to use for diffs with more lines than diff-threshold.
  • Default: gemini-2.5-flash
• diff-threshold: The line count threshold to switch between short-model and long-model.
  • Default: 500

Here is an example .autocommitmsg.yaml file:

base-url: "https://api.openai.com/v1"
api-key: "OPENAI_API_KEY"
short-model: "o4-mini"
long-model: "gpt-4.1"
diff-threshold: 250

Usage

After setting autocommitmsg as a prepare-commit-msg hook, invoking git commit without a commit message generates a commit message. If a commit message is given, autocommitmsg does not generate a commit message and instead uses the one provided by the user.

License

MIT