AI-generated conventional commit messages. Entirely on-device. No API keys. No code leaves your machine.
CommitCraft generates Conventional Commits from your staged git diff using a locally-running AI model (Qwen2.5-Coder-3B). The model runs entirely on your machine via a bundled llama.cpp binary — no internet connection required after the initial model download, no API keys, no telemetry.
Click the sparkle icon (✨) in the Source Control panel title bar, and CommitCraft fills the commit message box for you.
- One-click commit message generation from staged diffs
- Follows the Conventional Commits spec (
feat,fix,chore,refactor, etc.) - Respects your project's
commitlintconfiguration if present - 100% offline after first use
- Supports
commitlintscopes, breaking changes, and optional body generation - Works on Windows (x64), macOS (arm64 / x64), and Linux (x64)
- Install from the VS Code Marketplace
- On first use, CommitCraft will prompt you to download the AI model (~1.9 GB) from Hugging Face. This happens once and is cached in
~/.commitcraft/models/ - Stage some changes in git, click ✨ in the Source Control panel — done
CommitCraft is free to use for 14 days (trial). After the trial, a one-time license key is required.
Purchase at commitcraft.cyferlawyn.org — $9 one-time, perpetual.
- Open the Command Palette (
Ctrl+Shift+P/Cmd+Shift+P) - Run CommitCraft: Enter License Key
- Paste your key — validation is instant and fully offline
All settings are under commitcraft.* in VS Code settings:
| Setting | Default | Description |
|---|---|---|
commitcraft.modelPath |
(auto) | Override the model file path |
commitcraft.modelUrl |
(HuggingFace) | Override the model download URL |
commitcraft.inferenceTimeout |
30 |
Seconds to wait for inference |
commitcraft.maxDiffTokens |
4000 |
Max diff tokens sent to the model |
commitcraft.generateBody |
false |
Also generate a commit body paragraph |
macOS may quarantine the bundled llama.cpp binary on first run. If you see a security warning, run once in terminal:
xattr -dr com.apple.quarantine ~/.vscode/extensions/cyferlawyn.cyferlawyn-commitcraft-*/bin/darwin-*/Then reload VS Code.
- Your code never leaves your machine
- No analytics, no telemetry, no network calls during inference
- License validation is cryptographic (Ed25519) and fully offline
- The model is downloaded once from Hugging Face over HTTPS and verified by SHA-256
The extension source is open under MIT for the source code. Pre-built binaries are covered by a separate commercial license. See LICENSE for details.
GitHub: cyferlawyn/commitcraft-vscode