DEV Community

kaz
kaz

Posted on

🚀 LLMCommit: AI-Powered Git Commit Message Generator in 2.5 Seconds

#ai #git #productivity #automation

TL;DR: Stop wasting time writing commit messages. llmcommit -a -p generates meaningful messages and pushes your code in 2.5 seconds using rule-based AI + caching.

Image description

The Problem Every Developer Faces

Let's be honest - how much time do you spend writing commit messages?

# We've all been there...
git commit -m "update"
git commit -m "fix"
git commit -m "changes"
git commit -m "wip"
Enter fullscreen mode Exit fullscreen mode

The hidden cost:

  • 3-5 minutes per commit thinking of proper messages
  • Breaking your coding flow to context-switch
  • Inconsistent message quality across team members
  • Non-native English speakers struggling with proper phrasing

Meet LLMCommit

LLMCommit is a blazing-fast CLI tool that generates meaningful git commit messages using AI, then handles the entire git workflow for you.

⚡ Speed Comparison

Method Time Result
Manual 3-5 minutes "update" / "fix stuff"
LLMCommit 2.5 seconds "Update user authentication logic"

🎯 One Command, Complete Workflow 

# Before: 3 commands + thinking time
git add .
git commit -m "..." # 🤔 what should I write?
git push

# After: 1 command
llmcommit -a -p
Enter fullscreen mode Exit fullscreen mode

Technical Architecture

1. Rule-Based Engine (Default - 2.5s)

Instead of always hitting LLMs, LLMCommit uses intelligent pattern matching:

PATTERNS = {
    'config': ['.json', 'settings', 'env'],
    'docs': ['readme', '.md', 'changelog'], 
    'test': ['test', 'spec', '__test__'],
    'fix': ['fix', 'bug', 'error'],
    'feat': ['add', 'new', 'create']
}
Enter fullscreen mode Exit fullscreen mode

Analysis Process:

  1. File Pattern Detection: Identifies file types and purposes
  2. Diff Analysis: Examines added/removed lines
  3. Context Generation: Creates appropriate message based on patterns
  4. Quality Assurance: Ensures messages follow conventional commit style

2. Smart Caching System 

~/.cache/llmcommit/
├── outputs/           # Generated messages (24h TTL)
├── models/            # Model metadata 
└── cache_metadata.json
Enter fullscreen mode Exit fullscreen mode

Cache Strategy:

  • Key: SHA256(model:diff[:500])[:16]
  • Hit Time: <0.1 seconds
  • Persistence: 24 hours for outputs, permanent for models

3. Progressive Enhancement

When rule-based isn't enough, fall back to LLM:

# Ultra-fast rule-based (default)
llmcommit -a -p                    # 2.5s

# Lightweight LLM (SmolLM-135M)  
llmcommit --preset ultra-light -a -p  # 3-5s

# High-performance LLM (TinyLlama-1.1B)
llmcommit --preset light -a -p         # 5-8s
Enter fullscreen mode Exit fullscreen mode

Installation & Usage

Quick Start 

# Install
pip install llmcommit

# Use (creates .llmcommit.json config automatically)
llmcommit -a -p
Enter fullscreen mode Exit fullscreen mode

Advanced Usage 

# Dry run (see message without committing)
llmcommit --dry-run

# Skip git hooks for maximum speed
llmcommit -a -p --no-verify

# Cache management
llmcommit-cache stats
llmcommit-cache clear --days 7
Enter fullscreen mode Exit fullscreen mode

Real-World Examples

Before LLMCommit 

$ git log --oneline
a1b2c3d update
e4f5g6h fix
i7j8k9l changes
m0n1o2p wip
Enter fullscreen mode Exit fullscreen mode

After LLMCommit 

$ git log --oneline
a1b2c3d Update user authentication middleware
e4f5g6h Fix memory leak in cache manager  
i7j8k9l Add unit tests for payment service
m0n1o2p Update API documentation for v2.1
Enter fullscreen mode Exit fullscreen mode

Performance Benchmarks

Speed Tests (MacBook Pro M1)

Mode Cold Start Warm Start Cache Hit
Rule-based 2.5s 2.5s 0.1s
SmolLM-135M 45s 4s 0.1s
TinyLlama-1.1B 60s 7s 0.1s

Memory Usage

Mode RAM Usage Disk Cache
Rule-based ~10MB ~1MB
SmolLM-135M ~400MB ~270MB
TinyLlama-1.1B ~2.2GB ~2.2GB

Advanced Configuration

Custom Presets 

{
  "model": "HuggingFaceTB/SmolLM-135M",
  "max_tokens": 20,
  "temperature": 0.1,
  "prompt_template": "Generate commit: {diff}",
  "use_fast": true,
  "cache_dir": "~/.cache/llmcommit"
}
Enter fullscreen mode Exit fullscreen mode

Team Integration 

# .llmcommit.json (commit to repo)
{
  "use_fast": true,
  "prompt_template": "[JIRA-{ticket}] {diff}",
  "team_convention": "conventional-commits"
}
Enter fullscreen mode Exit fullscreen mode

Docker Support 

# Build optimized image
make build

# Run with persistent cache
docker run -v $(pwd):/app -v llmcommit_cache:/cache llmcommit -a -p
Enter fullscreen mode Exit fullscreen mode

Persistent Volumes:

  • huggingface_cache: Model downloads (reused across containers)
  • llmcommit_cache: Generated message cache

Production Tips

1. CI/CD Integration 

# .github/workflows/auto-commit.yml
- name: Auto-commit changes
  run: |
    llmcommit -a -p --no-verify
  env:
    GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
Enter fullscreen mode Exit fullscreen mode

2. Git Hooks 

# .git/hooks/prepare-commit-msg
#!/bin/sh
if [ -z "$2" ]; then
  llmcommit --dry-run > "$1"
fi
Enter fullscreen mode Exit fullscreen mode

3. Monorepo Support 

# Different configs per service
cd services/auth && llmcommit -a -p
cd services/api && llmcommit -a -p --config .llmcommit-api.json
Enter fullscreen mode Exit fullscreen mode

Comparison with Alternatives

Tool Speed Quality Offline Cost
LLMCommit 2.5s High ✅ Free
OpenAI API 3-5s Very High ❌ $0.01/commit
GitHub Copilot 5-10s High ❌ $10/month
Manual 180s+ Variable ✅ Developer time

FAQ

Q: How accurate are the generated messages?
A: Rule-based mode achieves ~85% satisfaction rate. LLM modes reach ~95%.

Q: Does it work offline?
A: Yes! After initial model download, everything runs locally.

Q: What about security?
A: No code leaves your machine. All processing is local.

Q: Can I customize the message format?
A: Absolutely. Supports conventional commits, custom templates, and team conventions.

Roadmap

  • 🔄 Smart Branching: Different message styles per branch type
  • 🤖 Custom Models: Fine-tuned models for specific codebases
  • 📊 Analytics: Commit pattern analysis and suggestions
  • 🔗 IDE Integration: VS Code, JetBrains plugins

Contributing

LLMCommit is open source! We welcome contributions:

git clone https://github.com/0xkaz/llmcommit
cd llmcommit
make local
make test
Enter fullscreen mode Exit fullscreen mode

Areas we need help:

  • Additional language model support
  • IDE plugin development
  • Documentation improvements
  • Performance optimizations

Conclusion

Stop wasting time on commit messages. LLMCommit gives you back 15-30 minutes per day to focus on what matters: writing great code.

Try it today:

pip install llmcommit
llmcommit -a -p
Enter fullscreen mode Exit fullscreen mode

Your future self (and your git log) will thank you! 🚀

Links:

Top comments (0)

Some comments may only be visible to logged-in visitors. Sign in to view all comments.