Git Commit Genie

Generates a commit message from your staged Git changes and writes it into the Source Control input box.

English | 简体中文

What it does

Git Commit Genie generates a commit message from the changes you have staged, using OpenAI, Anthropic Claude, Google Gemini or any OpenAI-compatible endpoint. Messages follow Conventional Commits 1.0.0 by default, and can follow a template kept in your repository.

Generation runs as a pipeline:

gitCommitGenie.generationMode decides how much work each commit gets and defaults to auto: Genie inspects the staged diff locally and writes changes it is confident about in a single pass, while everything else runs the full multi-stage pipeline. fast and deep pin one of the two — pick deep when every message must follow your template.

The pipeline is independent from the model's own reasoning level: the pipeline decides how many stages run, while gitCommitGenie.defaultThinkingLevel decides how much the model reasons inside each call.

Installation

Install from the VS Code Marketplace or Open VSX, or run:

code --install-extension Lune-99.git-commit-genie

Editors that use Open VSX instead of the Microsoft Marketplace, such as VSCodium, Cursor, Windsurf and Trae, can install the Open VSX build from their Extensions view.

Requirements: VS Code 1.103 or newer, the built-in Git extension, and an API key for the provider or endpoint you choose.

Usage

1. Run Git Commit Genie: Manage Models, pick a vendor such as OpenAI, Anthropic, DeepSeek or OpenCode Zen, choose one of its preset models and paste the API key once. The key is shared by every model you add from that vendor and is stored in VS Code SecretStorage, not in settings.json.
2. Stage your changes.
3. Click the Genie icon in the Source Control title bar, or run Git Commit Genie: Generate commit message.
4. Review the message in the commit box, edit it if needed, and commit.

Git Commit Genie: Menu opens the pipeline toggle, model management and repository memory.

Features

• Conventional Commits output, validated before it is written to the commit box.
• Providers: OpenAI, Anthropic Claude and Google Gemini, DeepSeek, GLM, Kimi, Qwen (international and China), and the OpenCode Zen and OpenCode Go gateways ship as presets with curated model lists and built-in pricing, plus custom OpenAI-compatible endpoints such as OpenRouter or local vLLM, SGLang and llama.cpp servers.
• Thinking levels from off to max, set globally or per model.
• Commit message languages: auto, en, zh-CN, zh-TW, ja, ko, de, fr, es, pt, ru, it.
• Extension UI in English, Simplified Chinese and Traditional Chinese.
• Generate and cancel from the Source Control title bar; progress is shown in the status bar.
• Optional repository investigation, repository memory and RAG style index (see Advanced features).
• Per-repository cost tracking (see Panel and cost).
• Templates for body structure, footers and vocabulary (see Templates).

Example output

feat(scm): add per-repository cost tracking to the Genie panel

- Record token usage and estimated cost for every generation
- Show accumulated spend next to each repository in the panel
- Keep the pipeline log available for the current session

Refs: [#142](https://github.com/Nouvelle-Lune/git-commit-genie/issues/142)

The structure depends on your template. Without a template, Genie follows Conventional Commits 1.0.0.

Panel and cost

The Genie panel in the Source Control sidebar lists the repositories in your workspace with their accumulated cost, and a log of pipeline stages and model calls.

While a message is generating, the status bar shows the current model and stage.

Templates

Templates are Markdown files stored in .gitgenie/templates in the workspace, or in your user template folder. Genie extracts the body layout, required footers and vocabulary preferences from the template and applies them while drafting and validating the message.

Guides: English and 中文.

Advanced features

Repository investigation is enabled by default. When the diff alone is ambiguous, the pipeline plans questions about the change and answers them from the repository: definitions, callers, configuration and tests. The number of lookups is bounded by gitCommitGenie.chain.investigation.maxSteps, and paths can be excluded with gitCommitGenie.chain.investigation.excludePatterns.

Repository memory is off by default. With gitCommitGenie.memory.enabled, Genie records what its investigations learned into local, per-repository storage and recalls it for similar changes later. Consolidation runs while the editor is idle. Use Git Commit Genie: Manage Repository Memory to inspect, rebuild or clear it.

RAG style index is off by default. With gitCommitGenie.rag.enabled, Genie indexes your historical commit messages locally and retrieves style references for new messages. It requires an OpenAI-compatible embeddings endpoint; configure it with Git Commit Genie: Configure RAG Embedding API Key.

Configuration

Commonly changed settings:

All commands and every setting are documented in the Commands & Settings reference.

Privacy

• API keys are stored in VS Code SecretStorage and are not written to settings.json or the logs.
• Only staged changes are analyzed. gitCommitGenie.autoStageAllForDiff is the exception; it restores your staging state afterwards.
• With investigation enabled, Genie reads files inside the repository and sends the findings it selects to the model provider you configured. Build outputs, dependency directories and lockfiles are excluded by default, and you can add exclusions.
• RAG sends historical commit messages to the embeddings endpoint you configure.
• Repository memory is stored in VS Code's local global storage and is not uploaded.
• The extension collects no analytics or telemetry.

Contributing

Issues and pull requests: https://github.com/Nouvelle-Lune/git-commit-genie.

License

MIT

Acknowledgements

• Conventional Commits
• OpenAI, Anthropic and Google Gemini, and the OpenAI-compatible model ecosystem.