mirror of
https://github.com/8bit-wraith/aye-commits.git
synced 2026-11-21 23:09:56 +01:00
Upstream pull mirror of https://github.com/8bit-wraith/aye-commits.git
- TypeScript 87.1%
- JavaScript 12.9%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| .github | ||
| .vscode | ||
| extension/assets/icons | ||
| public | ||
| src | ||
| .cspell.json | ||
| .gitignore | ||
| .npmrc | ||
| .vscode-test.mjs | ||
| .vscodeignore | ||
| CHANGELOG.md | ||
| esbuild.js | ||
| eslint.config.mjs | ||
| extensions.json | ||
| launch.json | ||
| LICENSE | ||
| package-lock.json | ||
| package.json | ||
| pnpm-lock.yaml | ||
| README.md | ||
| renovate.json | ||
| settings.json | ||
| tasks.json | ||
| tsconfig.json | ||
AI Commit Message Generator for VS Code 🤖✨
Local AI Commit Message Generation - Transform code diffs into conventional commits using open-source models. Privacy-focused, offline-capable solution for developers.
![]() |
![]() |
|---|
Demo showing commit message generation process (left) and settings (right)
Features 🌟
- 🔒 Privacy First - No data leaves your machine
- ⚡ Multi-Backend Support - Compatible with popular AI runners
- 📜 Commit Standard Compliance - Conventional Commits 1.0.0
- 🖥️ Hardware Aware - Optimized for various setups
- 🌐 Model Agnostic - Use any compatible LLM
Quick Start 🚀
- Install extension:
code --install-extension Its-Satyajit.ai-commit-message - Set up AI backend:
# For CPU-focused systems ollama pull phi-3 # For GPU-equipped machines ollama pull deepseek-r1:8b - Generate your first AI commit via VS Code Source Control view
Hardware Requirements 🖥️
Tested Environment
OS: openSUSE Tumbleweed
CPU: Intel i7-8750H (6c/12t @4.1GHz)
GPU: NVIDIA GTX 1050 Ti Mobile 4GB
RAM: 16GB DDR4
Storage: NVMe SSD
Minimum Recommendations
- CPU: 4-core (2015+)
- RAM: 8GB
- Storage: SSD
- Node: ^22
- Vscode: ^1.92.0
Model Compatibility 🧠
Performance Characteristics
| Model Family | Example Models | Speed* | Quality* | Use When... |
|---|---|---|---|---|
| Lightweight | phi-3, mistral |
22 t/s | ██▌ | Quick iterations |
| Balanced | llama3, qwen |
14 t/s | ███▎ | Daily development |
| Quality-Focused | deepseek-r1 |
7 t/s | ████▋ | Complex changes |
* Metrics from personal testing on mobile GTX 1050 Ti (Q4_K_M quantization)
Speed vs Quality Tradeoff
▲
│
Quality │.....█████ (deepseek-r1)
│...███ (llama3)
│.██▌ (phi-3)
└───────────────────▶ Time
Configuration ⚙️
Backend Setup
Option 1: Ollama (Simplest)
curl -fsSL https://ollama.com/install.sh | sh
ollama serve
* For more info, visit Ollama
Option 2: LM Studio (Advanced)
lmstudio serve --model ./models/deepseek-r1.gguf --gpulayers 20
* For more info, visit LM Studio
Extension Settings
{
"commitMessageGenerator.provider": "ollama",
"commitMessageGenerator.apiUrl": "http://localhost:11434",
"commitMessageGenerator.model": "deepseek-r1:8b",
"commitMessageGenerator.temperature": 0.7,
"commitMessageGenerator.maxTokens": 5000,
"commitMessageGenerator.apiKey": "your_api_key (if required by your OpenAI-compatible/ollama endpoint)",
"commitMessageGenerator.types": [
"feat: A new feature",
"fix: A bug fix",
"chore: Maintenance tasks",
"docs: Documentation updates",
],
"commitMessageGenerator.scopes": ["ui", "api", "config"]
}
Optimization Guide
GPU Acceleration
# NVIDIA Settings
export OLLAMA_GPUS=1
export GGML_CUDA_OFFLOAD=20
# Memory Allocation (4GB VRAM example)
┌───────────────────────┐
│ GPU Layers: 18/20 │
│ Batch Size: 128 │
│ Threads: 6 │
└───────────────────────┘
Performance Tips
- Start with
phi-3for quick feedback - Switch to
deepseek-r1for final commits - Use
--no-mmapif experiencing slowdowns - Reduce GPU layers when memory constrained
Troubleshooting 🔧
| Issue | First Steps | Advanced Fixes |
|---|---|---|
| Slow generation | 1. Check CPU usage 2. Verify quantization |
Use --no-mmap flag |
| Model loading fails | 1. Confirm SHA256 checksum 2. Check disk space |
Try different quantization |
| GPU not detected | 1. Verify drivers 2. Check CUDA version |
Set CUDA_VISIBLE_DEVICES=0 |
FAQ ❓
Why local AI instead of cloud services?
- Privacy: Code never leaves your machine
- Offline Use: Works without internet
- Cost: No API fees
- Customization: Use models tailored to your needs
How to choose between models?
Quick Sessions → phi-3/mistral:
- Prototyping
- Personal projects
- Low-resource machines
Important Commits → deepseek-r1:
- Production code
- Team projects
- Complex refactors
Legal & Ethics
Neutrality Statement
This project and I am not affiliated with, endorsed by, or sponsored by:
- Ollama
- LM Studio
- Any model creators
Mentioned tools/models are personal preferences based on technical merits.
Contributing 🤝
- Fork repository
- Install dependencies:
npm install - Build extension:
npm run package - Submit PR with changes
License 📄
MIT License - View License
Built by Developers, for Developers - From quick fixes to production-grade commits 💻⚡

