noidea

Because you shouldn't have to think about commit messages.

Python-native AI commit message generator — hooks into git commit, learns your repo's voice, routes small diffs to cheap models. No Node.js required.

---

Why noidea?

Most AI commit message tools require Node.js, generate the same boilerplate every time, and run as a separate command you have to remember to call. noidea is a Python-native alternative that works differently in the ways that actually matter:

• Writes in your repo's voice, not generic AI boilerplate. Before generating, noidea reads how your repo already commits — its scope vocabulary, body habits, subject length, gitmoji or not — and matches it. Your history stays consistent instead of looking like a bot dropped in.
• Spends pennies and milliseconds on trivial commits. Small diffs go to a fast, cheap model; only substantial changes escalate to the strong one — automatically, and it tells you which it chose. No paying premium rates to commit a typo fix.
• Never auto-commits — you always get the last word. It pre-fills your editor on git commit. No new command to remember, no message committed behind your back. Edit or delete it like any draft.
• Terminal-native and private by design. A small, MIT-licensed Python CLI built to a strict safety-first style guide. Your diffs go straight to your chosen model and nowhere else.

How noidea compares

	noidea	opencommit	aicommits	gptcommit
Install	pipx install noidea	npm i -g opencommit	npm i -g aicommits	cargo / brew
Language	Python	Node.js	Node.js	Rust
Hooks into git commit	✅	✅	✅	✅
Learns your repo's style	✅	❌	❌	❌
Cost-aware model routing	✅	❌	❌	❌
Multi-provider (6+)	✅	✅	✅	partial
No Node.js required	✅	❌	❌	✅
GitHub Action	✅	❌	❌	❌

Quick Start

pipx install noidea
noidea init

That's it. Every git commit now opens your editor with a suggested message pre-filled.

Requires pipx. Alternatively: pip install noidea

Works with Claude, GPT, and local Ollama models (plus Gemini, DeepSeek, Groq) — switching is a one-line config change, see Multi-provider. Want to try it with zero setup and no API key? Point it at a local Ollama model.

API Key Setup

By default noidea uses Anthropic (Claude). The key is looked up in three places, in order:

Method	Command
Keyring (recommended)	noidea keys add
Environment variable	export ANTHROPIC_API_KEY=sk-ant-...
.env file	ANTHROPIC_API_KEY=sk-ant-... in a .env file

Other providers follow the same pattern with their own env var (OPENAI_API_KEY, GEMINI_API_KEY, DEEPSEEK_API_KEY, GROQ_API_KEY) or noidea keys add <provider>.

Ollama needs no key. It runs locally, so once provider is set to ollama (below) you can generate commit messages with nothing else configured.

Multi-provider

Set provider in your config (see Config) to route every request to a different backend — no code change, no extra install:

{ "llm": { "provider": "ollama" } }

provider	Needs a key?	Default endpoint
anthropic (default)	yes	Anthropic SDK default
openai	yes	OpenAI SDK default
ollama	no	http://localhost:11434/v1
gemini	yes	Google's OpenAI-compatible endpoint
deepseek	yes	https://api.deepseek.com
groq	yes	https://api.groq.com/openai/v1

Set base_url to override the endpoint — e.g. point openai at a self-hosted vLLM server. Remember to also set small_model/large_model to models the provider actually serves (e.g. "small_model": "llama3.2" for Ollama).

Reasoning models: OpenAI's reasoning models (the o1/o3/gpt-5 family) are supported. They take max_completion_tokens instead of max_tokens and ignore temperature; noidea maps the parameters automatically and, for any provider, recovers if a model unexpectedly rejects max_tokens. Note that reasoning models spend part of the token budget on hidden reasoning, so a low max_tokens may leave no room for the message — raise llm.max_tokens if you get an empty-completion error.

Commands

Command	Description
noidea init	Install the prepare-commit-msg hook. Backs up any existing hook. Respects core.hooksPath.
noidea suggest	Generate a commit message from the staged diff and print it.
noidea status	Show current config, API key status, and hook installation.
noidea keys	Manage API keys in the system keyring (show / add / remove).
noidea test	Send a test message to Claude to verify connectivity.
noidea update	Upgrade noidea via pipx (falls back to pip).
noidea --version	Print the current version.

noidea suggest options

-F, --file TEXT    Write message to file instead of stdout (used by the hook)
-M, --model TEXT   Override the model used for generation

GitHub Action

noidea is also a GitHub Action. Add it to any repo and get AI-suggested commit messages posted as PR comments automatically.

# .github/workflows/noidea-suggest.yml
name: AI commit suggestion
on: [pull_request]

jobs:
  suggest:
    runs-on: ubuntu-latest
    permissions:
      pull-requests: write
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: AccursedGalaxy/noidea@main
        with:
          api-key: ${{ secrets.ANTHROPIC_API_KEY }}
          provider: anthropic        # or openai, groq, deepseek, gemini, ollama
          post-comment: 'true'

Switch to a free local model with provider: ollama and no api-key — no cost, no key rotation.

Config

Two optional config levels — both are config.json files:

• User: ~/.noidea/config.json — applies everywhere
• Repo: <repo>/.noidea/config.json — overrides user config

Precedence: built-in defaults → user config → repo config.

{
  "llm": {
    "max_tokens": 1024,
    "small_model": "claude-haiku-4-5",
    "large_model": "claude-sonnet-4-6",
    "context_limit": 600000,
    "temperature": 1.0,
    "learn_commit_style": true,
    "provider": "anthropic",
    "base_url": "",
    "system_prompt": "Your custom prompt here"
  }
}

Falls back to built-in defaults if no config file exists. The default prompt follows conventional commits style (feat/fix/refactor/etc.) with a 72-character subject line limit. Smaller diffs use small_model (Haiku) for speed; larger diffs automatically switch to large_model (Sonnet). temperature controls output creativity (0.0–1.0); the default of 1.0 maximises variety.

Repo-native style learning

learn_commit_style (on by default) samples your recent git log and matches the repo's own conventions — its scope vocabulary, whether commits carry a body, typical subject length, and gitmoji usage — so suggestions read like your project's existing commits rather than generic boilerplate. It activates automatically once a repo has at least 5 commits; below that it stays out of the way. Set "learn_commit_style": false to disable it.

Contributing

See CONTRIBUTING.md for development setup and guidelines. This project follows TigerStyle for coding standards.

Requirements

• Python 3.10+
• An LLM provider: an Anthropic, OpenAI, Gemini, DeepSeek, or Groq API key — or a local Ollama install (no key needed)