Commit Message Generator
Generate Conventional Commits messages from your staged diff using any OpenAI-compatible LLM endpoint.
Works with OpenAI, OpenRouter (free tier), Ollama, LM Studio, Groq, Together, vLLM, and more.
Features
- Sparkle button on the Source Control title bar — one click to generate.
- Streams tokens directly into the commit input box as they arrive.
- Any OpenAI-compatible endpoint — point at OpenRouter for free cloud models or Ollama for fully local, air-gapped generation.
- Conventional Commits format enforced by default (configurable).
- API keys stored securely in VS Code SecretStorage (OS keychain) — never in
settings.json.
- Diff truncation with per-file fair-share budget — large repos don't silently produce bad prompts.
Quick Start
- Install the extension.
- Open a git repository and stage some changes.
- Run Commit Gen: Set API Key from the Command Palette (or skip for Ollama / local endpoints).
- Click the ✨ button in the Source Control title bar (or run Commit Gen: Generate Commit Message).
Provider Recipes
OpenRouter (free tier)
"commitgen.baseUrl": "https://openrouter.ai/api/v1",
"commitgen.model": "deepseek/deepseek-chat-v3.1:free",
"commitgen.extraHeaders": {
"HTTP-Referer": "https://github.com/hirale/vscode-commitgen",
"X-Title": "Commit Message Generator"
}
Run Commit Gen: Set API Key and paste your OpenRouter API key (free account at openrouter.ai).
Ollama (local, no API key)
"commitgen.baseUrl": "http://localhost:11434/v1",
"commitgen.model": "qwen2.5-coder:7b"
No API key needed. ollama pull qwen2.5-coder:7b first.
LM Studio (local, no API key)
"commitgen.baseUrl": "http://localhost:1234/v1",
"commitgen.model": "your-loaded-model-id"
OpenAI
"commitgen.baseUrl": "https://api.openai.com/v1",
"commitgen.model": "gpt-4o-mini"
Configuration
| Setting |
Default |
Description |
commitgen.baseUrl |
https://api.openai.com/v1 |
OpenAI-compatible endpoint base URL |
commitgen.model |
gpt-4o-mini |
Model ID |
commitgen.language |
English |
Output language |
commitgen.conventionalCommits |
true |
Enforce Conventional Commits format |
commitgen.includeRecentCommits |
false |
Pass recent commit subjects as context |
commitgen.recentCommitsCount |
5 |
How many recent commits to include |
commitgen.maxDiffChars |
16000 |
Max diff chars to send (truncates beyond) |
commitgen.temperature |
0.2 |
Sampling temperature |
commitgen.maxOutputTokens |
512 |
Max tokens in generated message |
commitgen.stream |
true |
Stream tokens into input box |
commitgen.systemPromptOverride |
"" |
Custom system prompt (overrides built-in) |
commitgen.extraHeaders |
{} |
Extra HTTP headers (e.g., OpenRouter attribution) |
commitgen.requestTimeoutMs |
60000 |
Request timeout in ms |
Commands
| Command |
Description |
Commit Gen: Generate Commit Message |
Generate from staged diff |
Commit Gen: Set API Key |
Store API key in OS keychain |
Commit Gen: Select Model |
Browse or enter model ID |
Privacy
Your staged diff is sent to the configured endpoint. For proprietary or sensitive code, use a local endpoint (Ollama, LM Studio) so data never leaves your machine.
Troubleshooting
401 Unauthorized — Run Commit Gen: Set API Key and ensure the key matches the configured baseUrl.
No staged changes — Run git add <files> first, or use the Source Control panel to stage.
Diff truncated warning — Increase commitgen.maxDiffChars or stage fewer files at once.
Ollama times out — Increase commitgen.requestTimeoutMs (default 60s) if the model is slow to load.
Acknowledgements
Inspired by vscode-commitgen by WoongheeLee.
