@chof64/aicommit

AI-powered commit message generator. Reads your staged git diff, sends it to the opencode.ai zen chat completions API, and writes a conventional-commit message after a quick confirmation prompt.

Requirements

• Node.js 20+ (uses native fetch)
• git on PATH
• An opencode.ai API key exposed as OPENCODE_API_KEY

Install

npm i -g @chof64/aicommit

Configure

Export your opencode.ai API key in your shell rc:

export OPENCODE_API_KEY=<your-key>

Usage

Stage your changes as usual, then run aicommit:

git add .
aicommit

Add a hint to steer the message — useful for non-obvious diffs:

git add src/auth.ts
aicommit fix race in token refresh

Flags:

| Flag | Description | | ----------------- | -------------------------------------------------------- | | --dry-run | Print the generated message, do not commit. | | -v / --verbose| Echo verbose progress to stderr (network, retries). | | -V / --version| Print the version and exit. | | -h / --help | Print the help text and exit. |

You will always be asked to confirm before git commit runs. Press n (or N) to abort; anything else (including just hitting Enter) confirms.

How it works

1. Runs git diff --cached and aborts if nothing is staged.
2. Sends the diff (plus any hint) to https://opencode.ai/zen/v1/chat/completions with model big-pickle.
3. Asks the LLM for a single conventional-commit message (<type>: <description>).
4. Shows you the result, waits for Y/n, then runs git commit -m.

The full prompt sent to the model is in src/config.ts — see SYSTEM_PROMPT and USER_PROMPT_TAIL.

About

aicommit was created as a personal script to use AI in writing commit messages. It was a simple tool, and a way for me to learn scripting. It started as a Python script (with a shebang) on my Mac, but I found good use for it, so I migrated it to TypeScript and published it to npm.

It may not be as feature rich as other similar tools — still new, and under development — but it's the first project of mine on npm, and the first with (planned) CI/CD workflows that automate development, testing and deployment.

What's next

I'm planning to add a few features as time goes on:

• Customizable commit types — including support for the Angular convention
• Multiple AI models — flexibility in which model to use, with OpenAI API compatibility
• Optimizations — to the current version and the prompt
• Tests — coverage for the core flow

License

MIT