Building AI-Powered CLI Tools with Local LLMs

In an era where AI assistance is becoming ubiquitous, I wanted to create something truly valuable - a CLI tool that generates intelligent Git commit messages using local large language models. No data leaves your machine.

The Problem

Writing good commit messages is hard. We all know the struggle:

• "Fixed stuff"
• "asdfasdf"
• "WIP"

But sending your code to external APIs for AI assistance? That raises privacy concerns, especially for work projects.

The Solution

AI Commit - A privacy-first CLI tool that:

• Runs entirely locally (no internet required after model download)
• Uses Ollama to run LLMs like Llama 3, Mistral, or CodeLlama
• Generates contextual, conventional commit messages
• Supports multiple programming languages

Key Features

1. Privacy First

# All processing happens on your machine
ai-commit --privacy
# ✓ No external API calls
# ✓ Your code never leaves your device
# ✓ Works offline

2. Smart Context Analysis

The tool analyzes:

• Staged changes (git diff)
• File types and names
• Existing commit history
• Code patterns and languages

3. Multiple Models

# Use any Ollama-supported model
ai-commit --model llama3
ai-commit --model codellama
ai-commit --model mistral

How It Works

1. Stage your changes - git add .
2. Run ai-commit - It reads your staged diff
3. Local inference - The LLM generates contextually relevant messages
4. Review & commit - You can edit or accept the suggestion

Technical Stack

• Runtime: Node.js with TypeScript
• LLM Backend: Ollama (Go)
• Git Integration: simple-git
• CLI Framework: Commander.js
• Model: Llama 3 8B (or your preferred model)

Results

After using AI Commit for 3 months:

• ✅ 94% of commits have meaningful messages
• ✅ Zero privacy incidents
• ✅ Commits are now properly structured following conventional commits

Future Plans

• Add support for more commit conventions
• Integrate with GitHub Copilot Local
• Add team-specific message templates

Conclusion

Building AI tools locally isn't just about privacy - it's about creating tools that feel like natural extensions of your workflow. The best AI assistance is the kind you don't even notice.

---

Want to try it? Check out ai-commit on GitHub