commit-msg - Generate git commit messages with a local LLM in Docker

Runs Ollama + a small CPU-only model (default: qwen2.5:0.5b, ~350MB) in a Docker container. No GPU required.

Usage: git diff --cached | commit-msg # Message for staged changes git diff | commit-msg # Message for unstaged changes commit-msg # Auto-detects git diff git diff | commit-msg -m gemma2:2b # Use a different model

Options: -m, --model NAME Ollama model (default: qwen2.5:0.5b) -n, --dry-run Show input without calling the model -p, --pull Force re-download the model -c, --cleanup Stop container after use -k, --keep-alive Keep container running (default)

Common CPU-friendly models: qwen2.5:0.5b ~350MB - Fast, good quality qwen2.5:1.5b ~1GB - Better quality gemma2:2b ~1.6GB - Good quality phi3:3.8b ~2GB - Higher quality, slower

Notes: - First run pulls the model from Ollama (~350MB for default) - Model cached in Docker volume (ollama-commit-msg) for reuse - Container stays running between runs for speed - Stop: docker stop commit-msg-ollama