GPTComet: AI-Powered Git Commit Message Generator And Reviewer

简体中文

• GPTComet: AI-Powered Git Commit Message Generator And Reviewer
  • 💡 Overview
  • ✨ Features
  • ⬇️ Installation
  • 📕 Usage
  • 🔧 Setup
    • Configuration Methods
    • Managing Providers
    • Provider Setup Guide
      • OpenAI
      • Gemini
      • Claude/Anthropic
      • Vertex
      • Azure
      • Ollama
      • Other Supported Providers
    • Manual Provider Setup
  • ⌨️ Commands
  • 🪝 Git Hook
  • ⚙ Configuration
    • file_ignore
    • provider
    • output
    • Markdown theme
    • Supported languages
    • console
    • Sensitive data scanning
  • 🔦 Supported Keys
  • 📃 Example
    • Basic Usage
    • Enhanced Error Messages
    • Progress Indicators
  • 💻 Development
    • Requirements
    • Setup
    • Running Tests
      • Go Tests
      • Python Tests
    • Code Quality
      • Go
      • Python
    • Build
  • 📩 Contact
  • ☕️ Sponsor
  • 📜 License

💡 Overview

GPTComet is an AI-powered developer tool that streamlines your Git workflow and enhances code quality through automated commit message generation and intelligent code review.

✨ Features

This project leverages the power of large language models to automate repetitive tasks and improve the overall development process. The core features include:

• Automatic Commit Message Generation: GPTComet can generate commit messages based on the changes made in the code.
• Git Hook Integration: Install a prepare-commit-msg hook once and let git commit fill the message automatically.
• Intelligent Code Review: Get AI-powered code reviews with actionable feedback and suggestions.
• Progress Indicators: Optional verbose mode shows real-time progress for long-running operations.
• Support for Multiple Languages: GPTComet supports multiple languages, including English, Chinese and so on.
• Customizable Configuration: GPTComet allows users to customize the configuration to suit their needs, such llm model and prompt.
• Support for Rich Commit Messages: GPTComet supports rich commit messages, which include a title, summary, and detailed description.
• Support for Multiple Providers: GPTComet supports multiple providers, including OpenAI, Gemini, Claude/Anthropic, Vertex, Azure, AWS Bedrock, Ollama, and others.
• Support SVN and Git: GPTComet supports both SVN and Git repositories.

⬇️ Installation

To use GPTComet, you can install it with Homebrew:

brew install belingud/tap/gptcomet

To upgrade a Homebrew-managed installation, use:

brew upgrade gptcomet

You can also download from Github release, or by install scripts:

curl -sSL https://cdn.jsdelivr.net/gh/belingud/gptcomet@master/install.sh | bash

Windows:

irm https://cdn.jsdelivr.net/gh/belingud/gptcomet@master/install.ps1 | iex

If you want to install specific version, you can use the following script:

curl -sSL https://cdn.jsdelivr.net/gh/belingud/gptcomet@master/install.sh | bash -s -- -v 0.4.2

irm https://cdn.jsdelivr.net/gh/belingud/gptcomet@master/install.ps1 | iex -CommandArgs @("-v", "0.4.2")

If you prefer to run in python, you can install by pip directly, it packaged the binary files corresponding to the platform already.

pip install gptcomet

# Using pipx
pipx install gptcomet

# Using uv
uv tool install gptcomet
Resolved 1 package in 1.33s
Installed 1 package in 8ms
 + gptcomet==0.1.6
Installed 2 executables: gmsg, gptcomet

📕 Usage

To use gptcomet, follow these steps:

1. Install GPTComet: Install GPTComet through pypi.
2. Configure GPTComet: See Setup. Configure GPTComet with your api_key and other required keys like:

• provider: The provider of the language model (default openai).
• api_base: The base URL of the API (default https://api.openai.com/v1).
• api_key: The API key for the provider.
• model: The model used for generating commit messages (default gpt-6.1-sol).

3. Run GPTComet: Run GPTComet using the following command: gmsg commit.

If you are using openai provider, and finished set api_key, you can run gmsg commit directly.

🔧 Setup

Configuration Methods

1. Direct Configuration

   • Configure directly in ~/.config/gptcomet/gptcomet.yaml.

2. Interactive Setup

   • Use the gmsg provider new command for guided setup (gmsg newprovider is deprecated and will be removed in v2.14.0; see Deprecations).

3. Environment Variables

   • If api_key is not set in the configuration file, GPTComet falls back to the provider's conventional environment variable, such as OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, AZURE_API_KEY, or <PROVIDER>_API_KEY for other providers.
   • The configuration file takes precedence when both are set.

Managing Providers

Inspect the active provider and every known provider, switch the default, and verify a provider with one minimal request:

gmsg provider list     # active provider, model, API base, and API key source
gmsg provider use claude
gmsg provider test      # probe the active provider; or: gmsg provider test deepseek

gmsg provider list shows where each provider's API key comes from:

• config (sk-ab***) — set in gptcomet.yaml (masked in the output);
• env OPENAI_API_KEY — the config value is empty and the built-in environment variable is set;
• not required — local providers such as ollama and lmstudio;
• missing — no key found; commands that call the provider fail until one is set.

gmsg provider use <name> switches the default provider and creates a default configuration block for a built-in provider that does not have one yet. It never overwrites an existing block, and it warns when the API key or a provider-specific field (for example azure.deployment_name) is still missing. Configured custom providers keep working through the OpenAI-compatible fallback.

gmsg provider test [provider] sends exactly one small request without retries, using the provider's saved configuration, and reports the model, API base, key source, latency, and a short response. It exits non-zero when the request fails and prints an actionable error (invalid key, rate limit, provider outage, or network/proxy problem). It does not need a Git repository.

To add or reconfigure a provider interactively, use gmsg provider new (gmsg newprovider is deprecated and will be removed in v2.14.0; see Deprecations).

Provider Setup Guide

gmsg provider new

    Select Provider

  > 1. azure
    2. chatglm
    3. claude
    4. cohere
    5. deepseek
    6. gemini
    7. groq
    8. kimi
    9. mistral
    10. ollama
    11. openai
    12. openrouter
    13. sambanova
    14. silicon
    15. tongyi
    16. vertex
    17. xai
    18. Input Manually

    ↑/k up • ↓/j down • ? more

OpenAI

OpenAI api key page: https://platform.openai.com/api-keys

gmsg provider new

Selected provider: openai
Configure provider:

Previous inputs:
  Enter OpenAI API base: https://api.openai.com/v1
  Enter API key: sk-abc*********************************************
  Enter max tokens: 1024

Enter Enter model name (default: gpt-6.1-sol):
> gpt-6.1-sol


Provider openai configured successfully!

Gemini

Gemini api key page: https://aistudio.google.com/u/1/apikey

gmsg provider new
Selected provider: gemini
Configure provider:

Previous inputs:
  Enter Gemini API base: https://generativelanguage.googleapis.com/v1beta/models
  Enter API key: AIz************************************
  Enter max tokens: 1024

Enter Enter model name (default: gemini-3.8-flash):
> gemini-3.8-flash

Provider gemini already has a configuration. Do you want to overwrite it? (y/N): y

Provider gemini configured successfully!

Claude/Anthropic

I don't have an anthropic account yet, please see Anthropic console

Vertex

Vertex console page: https://console.cloud.google.com

gmsg provider new
Selected provider: vertex
Configure provider:

Previous inputs:
  Enter Vertex AI API Base URL: https://us-central1-aiplatform.googleapis.com/v1
  Enter API key: sk-awz*********************************************
  Enter location (e.g., us-central1): us-central1
  Enter max tokens: 1024
  Enter model name: gemini-3.8-flash

Enter Enter Google Cloud project ID:
> test-project


Provider vertex configured successfully!

Azure

gmsg provider new

Selected provider: azure
Configure provider:

Previous inputs:
  Enter Azure OpenAI endpoint: https://gptcomet.openai.azure.com
  Enter API key: ********************************
  Enter API version: 2024-02-15-preview
  Enter Azure OpenAI deployment name: gpt-6.1-sol
  Enter max tokens: 1024

Enter Enter deployment name (default: gpt-6.1-sol):
> gpt-6.1-sol


Provider azure configured successfully!

Ollama

gmsg provider new
Selected provider: ollama
Configure provider:

Previous inputs:
  Enter Ollama API Base URL: http://localhost:11434/api
  Enter max tokens: 1024

Enter Enter model name (default: qwen3.8:27b):
> qwen3.8:27b


Provider ollama configured successfully!

Other Supported Providers

• AI21
• AWS Bedrock
• Baidu Qianfan (ERNIE)
• ByteDance Doubao
• Cloudflare Workers AI
• Groq
• Mistral
• Tongyi/Qwen
• XAI
• Sambanova
• Silicon
• Deepseek
• ChatGLM
• KIMI
• LM Studio
• LongCat
• Cohere
• OpenRouter
• Hunyuan
• ModelScope
• MiniMax

Default models are refreshed periodically; see docs/provider-models.md for the current defaults and the last verification date.

AWS Bedrock uses the OpenAI-compatible Bedrock runtime endpoint with a Bedrock API key (set bedrock.api_key or AWS_BEARER_TOKEN_BEDROCK). IAM access-key/SigV4 authentication is not supported yet.

Manual Provider Setup

Or you can enter the provider name manually, and setup config manually.

gmsg provider new
You can either select one from the list or enter a custom provider name.
  ...
  vertex
> Input manually

Enter provider name: test
Enter OpenAI API Base URL [https://api.openai.com/v1]:
Enter model name [gpt-6.1-sol]:
Enter API key: ************************************
Enter max tokens [1024]:
[GPTComet] Provider test configured successfully.

This is also the escape hatch for any OpenAI-compatible service that GPTComet does not ship a preset for: enter a custom provider name, then set its api_base, model, and api_key. Advanced users can also set completion_path, answer_path, stream_answer_path, extra_headers, and extra_body for non-standard APIs.

Cloudflare Workers AI is a built-in provider now. Select cloudflare in gmsg provider new and enter your account ID, API token, and Workers AI model (for example @cf/openai/gpt-oss-120b); GPTComet builds the OpenAI-compatible endpoint for you. The old /ai/run + result.response manual workaround is no longer needed.

$ gmsg provider new

Selected provider: cloudflare
Configure provider:

Enter Cloudflare account ID: <account_id>
Enter Cloudflare API token: *************************************
Enter Workers AI model ID [@cf/openai/gpt-oss-120b]:
Enter max tokens [1024]:

Provider cloudflare configured successfully!

⌨️ Commands

The following are the available commands for GPTComet:

• gmsg config: Config manage commands group.
  • get <key>: Get the value of a configuration key.
  • list: List the entire configuration content.
  • reset: Reset the configuration to default values (optionally reset only the prompt section with --prompt).
  • set <key> <value>: Set a configuration value.
  • path: Get the configuration file path.
  • remove <key> [value]: Remove a configuration key or a value from a list. (List value only, like fileignore)
  • append <key> <value>: Append a value to a list configuration.(List value only, like fileignore)
  • keys: List all supported configuration keys.
  • validate: Validate the current configuration and report issues.
• gmsg commit: Generate commit message by changes/diff.
  • --svn: Generate commit message for svn.
  • --dry-run: Dry run the command without actually generating the commit message.
  • -y/--yes: Skip the confirmation prompt.
  • --no-verify: Skip git hooks verification, akin to using git commit --no-verify
  • --secrets: Sensitive data scan mode for this run: off, warn, redact, or block (default from config).
  • --repo: Path to the repository (default ".").
  • --answer-path: Override answer path
  • --api-base: Override API base URL
  • --api-key: Override API key
  • --completion-path: Override completion path
  • --frequency-penalty: Override frequency penalty
  • --max-tokens: Override maximum tokens
  • --model: Override model name
  • --provider: Override AI provider (openai/deepseek)
  • --proxy: Override proxy URL
  • --retries: Override retry count
  • --timeout: Override request timeout in seconds
  • --temperature: Override temperature
  • --top-p: Override top_p value
• gmsg generate: Generate a commit message from the staged diff and print only the message to stdout.
  • --repo: Path to the repository (default ".").
  • --rich: Generate a rich commit message with details.
  • --secrets: Sensitive data scan mode for this run: off, warn, redact, or block.
  • API override flags: --model, --api-base, --api-key, --timeout, and the other advanced flags.
• gmsg hook: Manage the Git prepare-commit-msg hook.
  • install [--repo] [--force]: Install the hook; --force backs up an existing hook first.
  • uninstall [--repo]: Remove a gptcomet-managed hook and restore a backup.
  • status [--repo]: Show the hooks directory, state, and backup.
• gmsg provider: Inspect, switch, test, and add LLM providers.
  • list: Show the active provider plus every provider's model and API key source.
  • new: Configure a new provider interactively (replaces gmsg newprovider).
  • use <name>: Switch the default provider; seeds defaults for a built-in provider without a config block.
  • test [provider]: Send one minimal request and report connectivity, auth, latency, and a short response.
• gmsg newprovider: Deprecated; use gmsg provider new. Kept for now; scheduled for removal in v2.14.0 (deprecation schedule).
• gmsg review: Review staged diff or pipe to gmsg review.
  • --svn: Get diff from svn.
  • --stream: Stream output as it arrives from the LLM.
  • --secrets: Sensitive data scan mode for this run: off, warn, redact, or block.
  • --repo: Path to the repository (default ".").
  • --answer-path: Override answer path
  • --api-base: Override API base URL
  • --api-key: Override API key
  • --completion-path: Override completion path
  • --frequency-penalty: Override frequency penalty
  • --max-tokens: Override maximum tokens
  • --model: Override model name
  • --provider: Override AI provider (openai/deepseek)
  • --proxy: Override proxy URL
  • --retries: Override retry count
  • --timeout: Override request timeout in seconds
  • --temperature: Override temperature
  • --top-p: Override top_p value

Global flags:

  -c, --config string   Config file path
  -d, --debug           Enable debug mode

🪝 Git Hook

Install the hook once per repository:

gmsg hook install
gmsg hook status

The managed prepare-commit-msg hook generates a message from the staged diff before the editor opens. It fails open: if gptcomet is missing, the configuration is invalid, there is no staged diff, or the API call fails or times out, the commit message file is left unchanged and the commit continues.

gmsg hook uninstall          # remove the managed hook
gmsg hook install --force    # back up and replace an existing hook

Behavior:

• Skips git commit -m / -F, merges, squash commits, and git commit --amend.
• Skips when the message file already has non-comment content. The all-comment template used by the first commit is still filled; a custom template with real content is left untouched.
• GMSG_SKIP=1 skips the hook entirely.
• GMSG_FORCE=1 overrides the explicit-message and existing-content skips.
• GMSG_BIN=/path/to/gmsg selects the binary; GMSG_CONFIG=/path/to/config.yaml selects the configuration file. gmsg hook install -c <config> embeds that config path in the hook.
• gmsg hook install --force backs up an existing hook to prepare-commit-msg.gptcomet-backup; gmsg hook uninstall restores it.
• The hooks directory is resolved with git rev-parse, so core.hooksPath and linked worktrees are respected.
• Generation is synchronous, so the editor opens after the model responds; the configured <provider>.timeout bounds the wait.
• git commit --no-verify does not skip prepare-commit-msg; use GMSG_SKIP=1 instead.

To preview the exact message without committing:

gmsg generate
gmsg generate --rich

generate writes only the message to stdout; status and token-usage output goes to stderr, so it is safe to use in scripts.

⚙ Configuration

Here's a summary of the main configuration keys:

Note: <provider> should be replaced with the actual provider name (e.g., openai, gemini, claude).

Some providers require specific keys, such as Vertex needing project ID, location, etc.

The configuration file for GPTComet is gptcomet.yaml. The file should contain the following keys:

output.translate_title is used to determine whether to translate the title of the commit message.

For example in output.lang: zh-cn, the title of the commit message is feat: Add new feature

If output.translate_title is set to true, the commit message will be translated to 功能：新增功能. Otherwise, the commit message will be translated to feat: 新增功能.

In some case you can set complation_path to empty string, like <provider>.completion_path: "", to use api_base endpoint directly.

file_ignore

The file to ignore when generating a commit. The default value is

- bun.lockb
- Cargo.lock
- composer.lock
- Gemfile.lock
- package-lock.json
- pnpm-lock.yaml
- poetry.lock
- yarn.lock
- pdm.lock
- Pipfile.lock
- "*.py[cod]"
- go.sum
- uv.lock

You can add more file_ignore by using the gmsg config append file_ignore <xxx> command. <xxx> uses gitignore-style patterns:

• *.lock matches a.lock and vendor/a.lock (a pattern without / matches at any depth).
• vendor/** matches everything under vendor/, and docs/*.md matches only direct children of docs/.
• /go.sum matches only the repository-root go.sum; go.sum matches it at any depth.
• build/ matches a directory named build and everything below it.
• !important.lock re-includes a file matched by an earlier pattern.

Example:

gmsg config append file_ignore "vendor/**"
gmsg config append file_ignore "!vendor/keep.go"

sensitive data scanning

Before any diff is sent to a provider, GPTComet scans the added lines for credentials with a lightweight engine loaded from the MIT-licensed gitleaks rule data (222 rules, vendored under internal/secretscan/rules/). It never changes your staged diff; only the text sent to the model can be rewritten.

sensitive:
  mode: warn      # off | warn | redact | block
  allow: []       # regexes matched against the secret or the full line

Sensitive file names (.env*, private key material, credentials.json, .npmrc, .pypirc, .netrc) are flagged even when no per-line regex matches. Silence known false positives with sensitive.allow, or a single line with a gitleaks:allow comment. The Git hook stays fail-open: in block mode it prints the warning but does not block git commit.

Override the mode for one command (all commands that send a diff):

gmsg commit --secrets redact
gmsg generate --secrets off
gmsg review --secrets block

Not detected: secrets hidden inside base64/hex blobs, git history, or archive contents — this is a pre-send guard, not a replacement for gitleaks/GitHub secret scanning in CI.

provider

The provider configuration of the language model.

The default provider is openai. You can inspect and switch it with gmsg provider list and gmsg provider use <name>, and verify the configuration with gmsg provider test.

Provider config just like:

provider: openai
openai:
    api_base: https://api.openai.com/v1
    api_key: YOUR_API_KEY
    model: gpt-6.1-sol
    retries: 2
    max_tokens: 1024
    temperature: 0.7
    top_p: 0.7
    frequency_penalty: 0
    extra_headers: {}
    answer_path: choices.0.message.content
    completion_path: /chat/completions

If you are using openai, just leave the api_base as default. Set your api_key in the config section.

If you are using an openai class provider, or a provider compatible interface, you can set the provider to openai. And set your custom api_base, api_key and model.

For example:

Openrouter providers api interface compatible with openai, you can set provider to openai and set api_base to https://openrouter.ai/api/v1, api_key to your api key from keys page and model to openai/gpt-6.1-sol or some other you prefer.

gmsg config set openai.api_base https://openrouter.ai/api/v1
gmsg config set openai.api_key YOUR_API_KEY
gmsg config set openai.model openai/gpt-6.1-sol
gmsg config set openai.max_tokens 1024

Silicon providers the similar interface with openrouter, so you can set provider to openai and set api_base to https://api.siliconflow.cn/v1.

Note that max tokens may vary, and will return an error if it is too large.

output

The output configuration of the commit message.

The default output is

output:
    lang: en
    rich_template: "<title>:<summary>\n\n<detail>"
    translate_title: false
    review_lang: "en"
    markdown_theme: "auto"

You can set rich_template to change the template of the rich commit message, and set lang to change the language of the commit message.

Markdown theme

Supported markdown theme:

• auto: Auto detect markdown theme (default).
• ascii: ASCII style.
• dark: Dark theme.
• dracula: Dracula theme.
• light: Light theme.
• tokyo-night: Tokyo Night theme.
• notty: Notty style, no render.
• pink: Pink theme.

If you not set markdown_theme, the markdown theme will be auto detected. If you are using light terminal, the markdown theme will be dark, if you are using dark terminal, the markdown theme will be light.

GPTComet is using glamour to render markdown, you can preview the markdown theme in glamour preview.

Supported languages

output.lang and output.review_lang support the following languages:

• en: English
• zh-cn: Simplified Chinese
• zh-tw: Traditional Chinese
• fr: French
• vi: Vietnamese
• ja: Japanese
• ko: Korean
• ru: Russian
• tr: Turkish
• id: Indonesian
• th: Thai
• de: German
• es: Spanish
• pt: Portuguese
• it: Italian
• ar: Arabic
• hi: Hindi
• el: Greek
• pl: Polish
• nl: Dutch
• sv: Swedish
• fi: Finnish
• hu: Hungarian
• cs: Czech
• ro: Romanian
• bg: Bulgarian
• uk: Ukrainian
• he: Hebrew
• lt: Lithuanian
• la: Latin
• ca: Catalan
• sr: Serbian
• sl: Slovenian
• mk: Macedonian
• lv: Latvian

console

The console output config.

The default console is

console:
    verbose: true

When verbose is enabled (true), GPTComet provides enhanced user experience:

• Progress Indicators: Shows real-time progress for commit message generation and code review

  [1/2] Fetching git diff...
  ✓ Fetching git diff (0.07s)
  Discovered provider: mistral, model: codestral-latest
  [2/2] Generating message...
  ✓ Generating message (13.24s)
  

• Detailed Operation Information: Displays which provider and model are being used

• Enhanced Error Messages: All errors include:

  • Clear problem description
  • Specific suggestions for resolution
  • Relevant documentation links
  • Appropriate emoji indicators for quick identification

When verbose is disabled (false), GPTComet runs in silent mode with minimal output, suitable for scripting and automated workflows.

🔦 Supported Keys

You can use gmsg config keys to check supported keys.

📃 Example

Here is an example of how to use GPTComet:

Basic Usage

1. When you first set your OpenAI KEY by gmsg config set openai.api_key YOUR_API_KEY, it will generate config file at ~/.local/gptcomet/gptcomet.yaml, includes:

provider: "openai"
openai:
  api_base: "https://api.openai.com/v1"
  api_key: "YOUR_API_KEY"
  model: "gpt-6.1-sol"
  retries: 2
output:
  lang: "en"

2. Run the following command to generate a commit message: gmsg commit
3. GPTComet will generate a commit message based on the changes made in the code and display it in the console.

Enhanced Error Messages

GPTComet provides helpful error messages with actionable suggestions:

$ gmsg commit

❌ API Key Not Configured

Provider 'openai' requires an API key, but none was found.

What to do:
  • Set API key: gmsg config set openai.api_key <your-key>
  • Or set env var: export OPENAI_API_KEY=<your-key>
  • Check provider: gmsg config get openai

Docs: https://github.com/belingud/gptcomet#configuration

Progress Indicators

When console.verbose is enabled (default), you'll see real-time progress:

$ gmsg commit

[1/2] Fetching git diff...(0.07s)
Discovered provider: mistral, model: codestral-latest
[2/2] Generating message...
📤 Sending request to mistral...
Token usage> prompt: 1341, completion: 10, total: 1,351
✓ Generating message (13.24s)

feat: add user authentication feature

To disable progress indicators and run in silent mode:

gmsg config set console.verbose false

Note: Replace YOUR_API_KEY with your actual API key for the provider.

💻 Development

Requirements

• Go 1.25+
• Python 3.9+
• just command runner
• pytest (for Python tests)

Setup

If you'd like to contribute to GPTComet, feel free to fork this project and submit a pull request.

First, fork the project and clone your repo.

git clone https://github.com/<yourname>/gptcomet

Second, make sure you have uv, you can install by pip, brew or other way in their installation docs

Use just command to install dependencies:

just install

Running Tests

Go Tests

# Run all Go tests
go test ./...

# Run specific package tests
go test ./internal/llm/

# Run with coverage
go test -coverprofile=coverage.out ./...
go tool cover -html=coverage.out

# Using just
just test              # Run tests with coverage
just test-coverage     # Generate coverage report
just test-cover-func   # Show coverage by function

Python Tests

# Run Python wrapper tests
just test-py

# Run with coverage
just test-py-cov

# Or manually with uv
uv run pytest tests/py_tests/ -v
uv run pytest tests/py_tests/ --cov=py/gptcomet --cov-report=html

Code Quality

Go

# Static analysis
go vet ./...
staticcheck ./...

# Using just
just check             # Run go vet and staticcheck
just format            # Format Go code

Python

# Code linting
ruff check py/

# Formatting
ruff format py/

Build

# Build Go binary
just build

# Build all platforms
just build-all

# Build Python wheel
just build-py

📩 Contact

If you have any questions or suggestions, feel free to contact.

☕️ Sponsor

If you like GPTComet, you can buy me a coffee to support me. Any support can help the project go further.

Buy Me A Coffee

📜 License

GPTComet is licensed under the MIT License.