CommitBrief：本地 LLM 代码审查 CLI，支持 git diff 与 GitHub PR

Local LLM code review CLI for git diffs — review staged/unstaged changes, any git diff range, or a GitHub PR, right in your terminal.

Run a "second pair of eyes" review on your staged changes, a specific file, a single commit, or a PR-style three-dot range — without leaving the terminal. Output is rendered as colored markdown in the terminal, plain markdown to a file, or strict JSON for tooling — your choice.

Why

A real reviewer is the gold standard, but they aren't always available the moment you stage a change. CommitBrief gives you a quick, structured read on your diff before another human (or your future self) sees it — local-first (review output stays on your machine; the diff goes only to the provider you pick, or nowhere at all with Ollama), provider-agnostic, cache aware on unchanged diffs, and steerable with a repo's own COMMITBRIEF.md review rules.

Quick Start

Install CommitBrief with one of the methods under Installation, then:

commitbrief setup   # pick a provider, paste your API key, run a ping
commitbrief init    # write a starter COMMITBRIEF.md
commitbrief         # review your staged changes

Walkthrough with sample output: Run your first review.

Installation

brew install CommitBrief/tap/commitbrief                          # macOS / Linux
scoop bucket add commitbrief https://github.com/CommitBrief/scoop-bucket
scoop install commitbrief                                         # Windows
go install github.com/CommitBrief/commitbrief/cmd/commitbrief@latest

Pre-built binaries are attached to each tagged release. Full install paths, package-manager setup, and commitbrief upgrade: see Installation.

Usage

commitbrief                                # review your staged changes
commitbrief diff HEAD                      # review working tree vs HEAD
commitbrief diff main...feature/x          # review a PR
commitbrief --unstaged --dir app/Models    # narrow any scope to a directory

Shipped providers: anthropic, openai, gemini, ollama (native API/local), deepseek, mistral, cohere (OpenAI-compatible), and claude-cli, gemini-cli, codex-cli (reuse a local CLI subscription, no extra API key). Models, context windows, and per-token pricing: Providers.

Documentation

The full documentation is the source of truth for everything below the front door:

Which command do I need? — decision table across the whole command surface
Output formats — Cards, JSON schema, Markdown
Review pull requests and Gate CI on findings
MCP server — let an AI agent self-review
Policy gate — declarative .commitbrief/policy.yml
Configuration and Filtering scope
Credential audit — leaks
FAQ / Troubleshooting

Project Status

Stable. The stable line is an API freeze: CLI flag surface, the JSON schema (--json), COMMITBRIEF.md/OUTPUT.md formats, and the public config keys follow strict semver — breaking changes only ship in the next major line. Details: Compatibility. Upgrading across a major version? See the migration notes in CHANGELOG.md.

Contributing

See CONTRIBUTING.md for the build/test flow, and the org-wide guide for licensing and PR conventions. Bug reports and questions are welcome in Issues and Discussions.

Security

Do not report vulnerabilities in public issues. Use private vulnerability reporting; the full process is in the security policy.

License

GPL-3.0-or-later. Provider SDK dependencies are Apache-2.0 or MIT; the full audit is make license-check.