ollama-commit 0.1.0

Generate Git commit messages using local LLMs via Ollama

ollama-commit

Generate Git commit messages using local LLMs via Ollama — no cloud, no API keys, no data leaving your machine.

$ git add src/auth.py
$ ollama-commit

Staged files (1): src/auth.py
Querying mistral via http://localhost:11434…

╭─ Suggested commit message ──────────────────────────────╮
│ feat(auth): add JWT refresh token rotation               │
╰──────────────────────────────────────────────────────────╯

Commit with this message? [Y/n]:

Features

• Local-first — uses Ollama, everything runs on your machine
• Conventional Commits — output follows the Conventional Commits spec
• Multiple suggestions — get up to 5 variants and pick the best one (-n 3)
• Context hints — give the model a nudge (--hint "refactoring auth")
• Git hook — auto-fill the commit message editor on every git commit
• Interactive or silent — --yes flag for scripting and CI

Requirements

• Python ≥ 3.10
• Ollama running locally (ollama serve)
• At least one model pulled: ollama pull mistral

Installation

pip install ollama-commit

Or for development:

git clone https://github.com/dennisreichenberg/ollama-commit
cd ollama-commit
pip install -e ".[dev]"

Usage

Basic

# Stage your changes, then:
git add .
ollama-commit

Get multiple suggestions

ollama-commit --count 3

Use a different model

ollama-commit --model llama3
ollama-commit --model codellama

Add a context hint

ollama-commit --hint "this fixes the login redirect bug"

Auto-commit without prompting

ollama-commit --yes

Dry run (print only, do not commit)

ollama-commit --dry-run

List available models

ollama-commit models

Git hook (auto-fill on git commit)

# Install hook in current repo
ollama-commit install-hook

# Remove hook
ollama-commit uninstall-hook

Options

Option	Short	Default	Description
--model	-m	mistral	Ollama model to use
--host		http://localhost:11434	Ollama base URL
--count	-n	1	Number of suggestions (1–5)
--hint	-h		Optional hint for the model
--yes	-y		Auto-accept and commit
--dry-run			Print message, don't commit

Recommended models

Model	Size	Notes
mistral	4.1 GB	Best quality/speed for commits
llama3	4.7 GB	Excellent for English commits
codellama	3.8 GB	Code-focused, good for technical diffs
phi3	2.2 GB	Fast, good quality for small models
qwen2.5-coder	4.7 GB	Strong code understanding

How it works

1. Reads git diff --staged from the current repository
2. Sends the diff (and optional hint) to your local Ollama model
3. Receives a Conventional Commit formatted message
4. Optionally commits with that message after confirmation

Configuration via environment variables

export OLLAMA_COMMIT_MODEL=llama3
export OLLAMA_COMMIT_HOST=http://localhost:11434

(Env var support coming in v0.2)

License

MIT — see LICENSE

Contributing

PRs welcome! See CONTRIBUTING.md for guidelines.