Commit messages written
by local AI
CommitCraft reads your staged diff and generates a conventional commit message using an on-device model. No API keys. No internet. No code ever leaves your machine.
Why CommitCraft
Everything local. Nothing compromised.
Fully offline
The AI model runs entirely on your CPU. Your code never touches a remote server — not even for license validation.
Fast enough
Qwen2.5-Coder 3B generates a commit message in 2–6 seconds on a typical laptop CPU. No GPU required.
Conventional commits
Outputs type(scope): subject format every time. Respects your commitlint config when present.
One-time purchase
Pay once, use forever. No subscription, no seat limits, no expiry. License validated offline via Ed25519 signature.
Zero config
Install the extension, click the sparkle icon in the Source Control panel. The model downloads itself on first use.
Open source
Extension source is MIT-licensed on GitHub. Audit the code, verify there's no telemetry, build it yourself.
How it works
Three steps, no surprises
Install & download model
Install CommitCraft from the VS Code Marketplace. On first use it downloads the Qwen2.5-Coder 3B GGUF model (~1.9 GB) to ~/.commitcraft/models/ — once, then never again.
Stage your changes
Stage your files as normal with git add or the VS Code Source Control panel.
Click the sparkle
Hit the ✦ button in the Source Control title bar. CommitCraft reads the diff, runs the model locally, and writes the commit message directly into the input box.
Pricing
Simple, one-time pricing
CommitCraft License
one-time · no subscription · no expiry
- ✓ Unlimited generations
- ✓ All future updates
- ✓ Works on Windows, macOS, Linux
- ✓ Offline license validation
- ✓ License key emailed instantly
Free 14-day trial — no license needed to try it.