English | 简体中文

WTF Commit ✨

Links: GitHub | Open VSX | Website

One shortcut to generate your commit message. WTF Commit reads your git diff and fills the Source Control box with a clear Conventional Commit — with your own API key, not a locked-in vendor.

Works in Cursor, VSCodium, and other Open VSX–compatible editors (and VS Code when the extension is installed from Open VSX / VSIX). Actively maintained (MIT, free).

Onboarding is intentionally two phases. You only configure AI once; after that it should feel like a muscle-memory shortcut.

① Configure AI (once)

You need a provider and a key. Everything else can stay on defaults.

1. Install — Extensions → search WTF Commit (Open VSX) → Install.
2. Set API Key — Command Palette → WTF Commit: Set API Key.
3. Choose a provider in the picker — the extension default is DeepSeek (also great: Gemini). Leave Model empty for the built-in default.
4. Paste the key. If you picked someone other than DeepSeek, choose Switch Provider when prompted so the active provider matches the key.

Need a key? DeepSeek · Gemini · more in Supported providers.

You can also set Provider under Settings → WTF Commit, then run Set API Key for that provider. Same result.

② Daily use (every commit)

1. Make your code changes (stage when you can; with Auto Commit + Smart Stage, unstaged work can be staged for you).
2. Press the generate shortcut — the message streams in, then commits by default (no extra “confirm commit” dialog).
3. Push yourself when ready — Auto Push is off by default so new users never surprise-push.

Default shortcut: Cmd+Alt+G (Mac) / Ctrl+Alt+G (Windows/Linux).

Make it yours — the binding is fully customizable. A popular Cursor-style habit is a double press of Cmd+G / Ctrl+G (chord):

1. Open Keyboard Shortcuts (Cmd+K Cmd+S / Ctrl+K Ctrl+S).
2. Search WTF Commit: Generate.
3. Double-click the keybinding → press Cmd+G twice (or Ctrl+G twice on Windows/Linux) → Enter.

One-keystroke power flow (optional): Settings → enable Auto Push. Leave Confirm Before Push on until you trust it; turn that confirm off only when you want generate → commit → push with zero dialogs.

Other triggers: ✨ on the Source Control title bar, or Command Palette → WTF Commit: Generate.

Prefer review-only? Turn Auto Commit off — the message stays in Source Control for you to edit and commit manually.

🆕 Latest (v1.13.0)

• Default provider: DeepSeek — first install uses deepseek-v4-flash (fast, low-cost commit messages).
• Smoother defaults: Auto Commit on, no confirm-before-commit dialog; Auto Push stays off; Confirm Before Push stays on for safety.
• Clearer onboarding docs: configure AI once, then daily shortcut (including rebinding tips like double Cmd+G).

See CHANGELOG for earlier releases.

🚀 Features

• Conventional Commits — feat / fix / docs / … with optional local format fix + AI Repair.
• Smart diffing — Prefers staged changes; confirms mixed or working-tree-only cases so the message matches what you intend to commit.
• Intent-aware — Text already in the SCM input is used as a generation hint (no extra prompt UI).
• Streaming preview — Watch the message appear live while the model runs.
• Auto Commit & Push — Optional one-keystroke pipeline with confirmations.
• Multi-language messages — English, 简体/繁体中文, Japanese, Classical Chinese, or Custom.
• Bring your own endpoint — Built-in providers plus Custom (Ollama, proxies, etc.).
• Keyboard shortcut — Default Cmd+Alt+G / Ctrl+Alt+G; rebind freely (e.g. double Cmd+G).

🛠️ Advanced Tutorial

1. Plugin Settings

Open VS Code Settings (Cmd+,) and search for WTF Commit to customize the behavior:

2. Custom Model & Endpoints

You can use any OpenAI-compatible model (like local models via Ollama) by changing the Provider or Model/Base URL:

1. In Settings, set Provider to Custom.
2. Enter the Base URL (e.g., http://localhost:11434/v1).
3. Enter the Model name (e.g., llama3).

3. Custom Commit Language

If you want the AI to use a specific language (e.g., French, Cantonese, or Emoji-only):

1. Set Language to Custom.
2. Enter your target language in Custom Language (e.g., Emoji only).

ℹ️ Supported Providers & Models

How defaults work (three separate concepts):

Provider defaults — when Base URL and Model are left empty:

Get API Keys

GLM and Z.AI keys are not interchangeable — create a key on the platform that matches your provider.

Choosing a model for commit messages

Generating a commit message is a lightweight task — you don't need a frontier model. Pick based on cost, latency, and whether you already have an API key.

The list below is our recommendation — change Provider in settings to use a different service (e.g. set Provider to DeepSeek and leave Model empty for deepseek-v4-flash).

Pricing comparison (USD per 1M tokens, cache-miss input; sources linked below):

† Rough estimate for ~5K input + 150 output tokens (typical diff + commit message), no prompt cache. GLM CNY prices converted at ~¥7.2/$ for comparison. Actual cost depends on diff size and model verbosity.

GLM vs Z.AI: Same model family, different platforms — GLM uses the China endpoint (open.bigmodel.cn); Z.AI uses the international endpoint (api.z.ai). API keys are not interchangeable. Each uses glm-4.7-flashx as its provider default (paid). The free glm-4.7-flash tier is heavily rate-limited and not recommended.

Our recommendation (speed + value) — set Provider in settings to match your pick:

1. DeepSeek V4 Flash — Default Provider; leave Model empty → deepseek-v4-flash. Best overall: fast, cheap, high quality. Thinking mode is disabled automatically.
2. Gemini 3.1 Flash Lite — set Provider to Gemini; leave Model empty → gemini-3.1-flash-lite. Also excellent: fast with a generous free tier; uses thinking_level: minimal.
3. MiMo V2.5 — set Provider to MiMo; leave Model empty → mimo-v2.5. Same USD price band as DeepSeek.
4. GLM / Z.AI — set Provider to GLM (China) or Z.AI (international); leave Model empty → glm-4.7-flashx. Works reliably but is often slower than DeepSeek or Gemini Flash Lite. Requires a paid balance.
5. openrouter/free — set Provider to OpenRouter; fine for experimenting.
6. NVIDIA NIM — set Provider to NVIDIA NIM; leave Model empty → nvidia/nemotron-3-super-120b-a12b. Good for free development testing across NVIDIA's hosted model catalog; expect lower rate limits and no production SLA.

Official pricing pages: DeepSeek · MiMo · Gemini · OpenAI · Zhipu GLM · Z.AI · OpenRouter · NVIDIA NIM

OpenRouter provider default is openrouter/free.

NVIDIA NIM provider default is nvidia/nemotron-3-super-120b-a12b. NVIDIA NIM is best treated as a free development/testing endpoint, not a guaranteed production backend.

Gemini uses Google's native Interactions REST API (/v1beta/interactions), authenticates with the x-goog-api-key header, and uses the minimal thinking level to reduce latency for commit-message generation.

[!IMPORTANT] Claude Support: Native Claude format is not supported yet. Please use a proxy service that provides an OpenAI-compatible endpoint.

🕹️ Other entry points

• Source Control title bar ✨ · Command Palette WTF Commit: Generate · rebindable shortcut (see Daily use).
• Rotate keys anytime with WTF Commit: Set API Key.

💬 Feedback

Using WTF Commit and finding it useful? A GitHub star, an Open VSX review, or a short issue (bugs or ideas) all help more people discover it — and help us keep improving.

📄 License

MIT License.