CommitLoom
Generate meaningful git commit messages with AI — in one click, right from Source Control. Supports NVIDIA and OpenRouter, with secret scanning built in.
Recommended: OpenRouter for the most reliable commit-message generation experience in CommitLoom.
Features
- ✨ One-click generation from the Source Control title bar or Command Palette
- 🤖 Two providers: NVIDIA (rated model picker) and OpenRouter (400+ models, free tier available)
- 📝 Conventional, simple, gitmoji, or fully custom commit styles
- 🔒 Diffs are scanned for secrets before anything leaves your machine
- ✍️ Messages auto-fill the commit box — never auto-committed, you always review first
Install
From the VS Code Marketplace (once published): open the Extensions view (Ctrl+Shift+X), search CommitLoom, click Install.
From a .vsix file: Extensions view → … menu → Install from VSIX… → pick commitloom-0.1.1.vsix. Or from a terminal:
code --install-extension commitloom-0.1.1.vsix
Get an API key
You need one key — either NVIDIA or OpenRouter. Keys are stored in VS Code SecretStorage, never in settings.json.
OpenRouter (recommended, free tier available)
- Go to openrouter.ai and sign up / sign in.
- Open the Keys page: openrouter.ai/keys.
- Click Create Key, name it (e.g.
CommitLoom), and copy it — it's shown only once.
- Back in VS Code:
Ctrl+Shift+P → CommitLoom: Configure Provider → OpenRouter → paste the key.
- Pick a model (default: Free Models Router — no credits needed) or enter any
provider/model ID from openrouter.ai/models.
- Optional: paid models need credits — top up at openrouter.ai/credits.
NVIDIA (free to start)
- Go to build.nvidia.com and sign in (or create a free NVIDIA account).
- Open your profile menu → API Keys (or visit
build.nvidia.com → keys section).
- Click Generate Key, give it a name, and copy it — it's shown only once.
- Back in VS Code:
Ctrl+Shift+P → CommitLoom: Configure Provider → NVIDIA → paste the key.
- Pick a model (recommended: DeepSeek V4 Pro) or enter a custom model ID.
How it works
- Stage your changes in Source Control (
git add, or stage in the UI).
- Press the CommitLoom button
in the Source Control title bar (top-right of the Source Control view) — or run CommitLoom: Generate Commit Message from the Command Palette.
- If nothing is staged, CommitLoom asks: generate from unstaged changes, stage everything, or cancel.
- Your diff is scanned for secrets first. If anything looks like a key or token, you choose: Redact and continue, Send anyway, or Cancel.
- The AI writes your commit message and drops it straight into the commit box. Review it, tweak if you like, then commit yourself. A notification offers Regenerate for another try.
Large diffs are truncated per file (largest first, lockfiles always summarized) and you'll be told which files were summarized.
Settings
| Setting |
Default |
Description |
commitloom.provider |
"nvidia" |
AI provider: nvidia or openrouter. Switch via Configure Provider. |
commitloom.model |
(provider default) |
Model ID. Change via Configure Provider (rated/catalog picker or custom ID). |
commitloom.commitStyle |
"conventional" |
conventional | simple | gitmoji | custom. |
commitloom.maxDiffSize |
30000 |
Max diff characters before per-file truncation kicks in. |
commitloom.customInstructions |
"" |
Extra prompt guidance (used as the style when commitStyle is custom). |
commitloom.model is shared between providers. Leave it empty to use the provider default. Switching providers via Configure Provider resets it before model selection, so cancelling setup still leaves a valid default.
NVIDIA troubleshooting
If NVIDIA returns an empty message, choose DeepSeek V4 Pro or Nemotron 3.5 Lightning via Configure Provider. CommitLoom disables thinking for these models so the output budget goes to the commit message. Other NVIDIA models receive a larger output budget, and NVIDIA requests can take up to 120 seconds. If a saved model is no longer available, select a current model from the picker.
Privacy & security
- Diffs are sent to the selected AI provider only to generate the message — nothing else.
- Secret scanning runs locally before any network call.
- API keys live in VS Code SecretStorage, never in settings files or the repo.
Future work
Local Ollama model support is coming soon. This planned integration will let you generate commit messages using models running on your own machine. Ollama support is not available in the current release.
Contributing
See CONTRIBUTING.md for build instructions and the provider architecture.
