GitMsgOllama

Install from VS Code Marketplace →

---

AI-powered commit message generator for VS Code. Get perfect commits in seconds using local AI with Ollama.

Why GitMsgOllama?

Better than writing commits yourself:

• Consistent format across your team
• Never forget what you changed
• Follows best practices automatically
• Saves 5+ minutes per day

Completely free and private:

• Zero API costs - runs entirely on your machine
• Complete privacy - your code never leaves your computer
• Offline use - works without internet
• No account required - no API keys needed

Quick Start (2 minutes)

1. Install Ollama (takes 1-2 minutes)
2. Run: ollama pull qwen2.5-coder
3. Install extension
4. Click the robot icon in your Git panel
5. Select "Local" provider
6. Done! No API key, no credit card, no tracking.

Features

Core Features

• AI-Powered Commit Messages: Uses AI to analyze your changes and generate meaningful commit messages
• Local AI Only: Privacy-first approach with Ollama for completely local operation
• Conventional Commits Support: Automatically follows conventional commit format with configurable types and scopes
• Smart Caching: Reuses recent commit messages for identical changes
• Review Mode: Preview and approve commit messages before applying
• Customizable: Configure AI model, prompt template, and conventional commit rules

Privacy Features

• Local Processing: Your code never leaves your machine
• No API Keys: No account or credentials required
• Offline Capable: Works without internet
• Open Source: Audit the code yourself

---

Local AI Setup

GitMsgOllama uses Ollama for completely local, private AI-powered commit messages.

Supported Models

Model	Size	RAM Required	Best For
qwen2.5-coder	~4GB	8GB	Code-focused, recommended
codellama:7b	3.8GB	8GB	Coding, fast responses
mistral:7b	4.1GB	8GB	General purpose, quality
llama2:13b	7.3GB	16GB	Higher quality
deepseek-coder:6.7b	3.8GB	8GB	Code-specific tasks

Setup Steps

1. Install Ollama from Ollama.ai

2. Download a model:

   ollama pull qwen2.5-coder
   # or
   ollama pull codellama:7b

3. Install the extension from VS Code Marketplace

4. Configure in VS Code:

   • Open Command Palette (Ctrl+Shift+P / Cmd+Shift+P)
   • Run: GitMsgOllama: Select Provider
   • Choose "Local"
   • Set base URL: http://localhost:11434/v1 (default)
   • Run: GitMsgOllama: Select Model
   • Choose your downloaded model

5. Start generating commit messages!

For detailed setup instructions and troubleshooting, see docs/PROVIDERS.md.

Requirements

Hardware

• Minimum: 8GB RAM (for 7B models)
• Recommended: 16GB+ RAM (for 13B models)
• GPU: Optional but significantly faster (NVIDIA/AMD/Apple Silicon)

Software

• VS Code
• Ollama (running locally)

Usage

Basic Usage

1. Make changes to your code
2. Stage your changes in Git (using Source Control panel)
3. Click the robot head icon in the Source Control message input box or title bar
4. The AI will analyze your changes and generate a commit message
5. Edit the message if needed and commit as usual

Using Cache

When you generate a commit message for the same set of changes:

1. The extension checks if a cached message exists
2. If found, you'll see: "Found cached suggestion from X minutes ago"
3. Choose:
   • Use Cached - Apply the previously generated message
   • Generate New - Create a fresh message
   • Dismiss - Cancel the operation

To clear the cache:

• Command Palette → GitMsgOllama: Clear Cache

Review Mode

Enable review mode to preview messages before applying:

{
  "gitmsgollama.reviewBeforeApply": true
}

With review mode enabled:

1. Generate a commit message
2. Review the suggested message in a quick pick dialog
3. Choose to accept, edit, or regenerate

Using Conventional Commits

The extension automatically generates conventional commit messages. Example output:

feat: add user authentication
fix(api): handle null response from endpoint
docs: update installation instructions
refactor(utils): simplify date formatting logic

Configure conventional commits:

{
  "gitmsgollama.conventionalCommits.enabled": true,
  "gitmsgollama.conventionalCommits.types": ["feat", "fix", "docs", "style", "refactor", "test", "chore"],
  "gitmsgollama.conventionalCommits.enableScopeDetection": true
}

Extension Settings

Core Settings

Setting	Type	Default	Description
gitmsgollama.provider	string	local	AI provider (local only)
gitmsgollama.local.model	string	qwen2.5-coder	Local model (Ollama)
gitmsgollama.local.baseUrl	string	http://localhost:11434/v1	Ollama server URL
gitmsgollama.prompt	string	(see below)	Custom prompt template (use {changes} placeholder)
gitmsgollama.timeout	number	30	API request timeout in seconds

Feature Settings

Setting	Type	Default	Description
gitmsgollama.reviewBeforeApply	boolean	true	Review messages before applying
gitmsgollama.enableCache	boolean	true	Enable commit message caching
gitmsgollama.cacheSize	number	10	Maximum cached messages

Conventional Commits Settings

Setting	Type	Default	Description
gitmsgollama.conventionalCommits.enabled	boolean	true	Enable conventional commits support
gitmsgollama.conventionalCommits.types	array	["feat", "fix", "docs", ...]	Allowed commit types
gitmsgollama.conventionalCommits.scopes	array	[]	Allowed scopes (empty = any)
gitmsgollama.conventionalCommits.enableScopeDetection	boolean	true	Auto-detect scope from file paths
gitmsgollama.conventionalCommits.requireScope	boolean	false	Require scope in messages

Default Prompt Template

The default prompt generates conventional commit messages:

Given these staged changes:
{changes}

Generate a commit message that follows these rules:
1. Start with a type (feat/fix/docs)
2. Keep it under 50 characters
3. Use imperative mood

You can customize this in settings to match your team's commit message style.

Privacy

What Data Stays Local

When you generate a commit message with local AI:

• Git diff of your staged changes stays on your machine
• Your prompt template with the diff stays local
• No external servers - nothing is sent to the internet
• No API keys or accounts required

Complete Privacy

• Your code never leaves your machine
• No data is sent to external AI providers
• No telemetry or analytics
• No usage tracking
• Open source - audit the code yourself

Troubleshooting

"Connection refused"

• Ensure Ollama is running
• Check base URL matches: http://localhost:11434/v1
• Check firewall isn't blocking local connections

"Model not found"

• Verify model is downloaded: Run ollama list
• Check model name matches exactly
• Download a model: ollama pull qwen2.5-coder

"Response too slow"

• Use smaller model (7B instead of 13B)
• Enable GPU acceleration in Ollama settings
• Reduce context length
• Use quantized models

"Out of memory"

• Use smaller model
• Close other applications
• Increase system swap/page file

For more detailed troubleshooting, see docs/PROVIDERS.md.

Commands

Command	Description
GitMsgOllama: Generate Commit Message	Generate a commit message for staged changes
GitMsgOllama: Select Provider	Choose your AI provider (Local)
GitMsgOllama: Set API Key	Not required for Local (no API key needed)
GitMsgOllama: Select Model	Browse and select from available Ollama models
GitMsgOllama: Test Provider Connection	Test your Ollama configuration
GitMsgOllama: Clear Cache	Clear all cached commit messages

Development

Want to contribute? See development setup:

1. Clone the repository
2. Run npm install
3. Open in VS Code
4. Press F5 to start debugging
5. Make your changes and submit a PR

License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

• Report bugs: GitHub Issues
• Security vulnerabilities: See SECURITY.md
• Feature requests: GitHub Issues

Acknowledgments

This project is a fork of GitMsgAI by Chase Rich. The original project provided the foundation for this privacy-focused Ollama-only version.