CommitAI — Local AI-Powered Git Assistant Using Gemma 4

This is a submission for the Gemma 4 Challenge: Build with Gemma 4

What I Built

As developers, we repeatedly write commit messages, maintain changelogs, and switch context during development. I wanted to build a lightweight developer tool that could automate this workflow completely offline without relying on cloud APIs or external AI services.

So I built CommitAI — a local AI-powered Git assistant using Gemma 4 + Ollama.

The workflow is simple:

• Read staged Git diff
• Send it to a local Gemma 4 model
• Generate a clean Conventional Commit message
• Update changelog automatically
• Execute the Git commit

I also added Git hook integration so the workflow becomes:

git add .
git commit

and the AI handles the rest.

The project is built entirely in Python and runs locally on my MacBook Air M2 (8GB RAM).

Features

• AI-generated Conventional Commit messages
• Automatic changelog generation
• Git hook integration
• Fully local/offline workflow
• Rich CLI interface using the Rich library
• No external APIs
• Privacy-friendly local inference
• Uses staged Git diffs as context

Demo

Demo Video

Code

CommitAI Repo

How I Used Gemma 4

I used the gemma4:e2b model through Ollama for all AI-powered functionality inside CommitAI.

I specifically chose the E2B model because:

• it runs efficiently on local hardware
• lightweight enough for my MacBook Air M2 with 8GB RAM
• low-latency responses for CLI workflows
• strong enough reasoning for summarizing Git diffs
• ideal for privacy-focused offline tooling

Gemma 4 powers:

• commit message generation
• changelog summarization
• Git workflow automation

Architecture

Git Diff
   ↓
CommitAI
   ↓
Ollama API
   ↓
Gemma 4 (E2B)
   ↓
AI-generated commit/changelog
   ↓
Git commit execution

Tech Stack

• Python
• Ollama
• Gemma 4 (gemma4:e2b)
• Rich
• Git Hooks

Future Improvements

• PR description generation
• Multi-model support
• Better diff summarization
• VSCode extension
• Team changelog modes
• Interactive commit editing

One of my biggest goals with this project was proving that useful AI developer tooling can run fully locally without depending on cloud APIs.

Thanks for reading 🚀