Codex-Clone: Open-Source AI Code Generator

A free, open-source command-line tool that generates code from natural language prompts using CodeLlama or StarCoder models.

Features

• 🚀 Free & Open-Source - No API keys required
• 💻 Local Execution - Run completely offline
• 🎯 Multiple Languages - Python, JavaScript, Go, Rust, C++, etc.
• ⚡ GPU Acceleration - Faster generation with CUDA
• 💾 Save to File - Direct output to code files

Requirements

• Python 3.8+
• PyTorch
• Transformers library
• CUDA-capable GPU (optional, but recommended)

Installation

# Clone the repository
git clone https://github.com/yourusername/codex-clone.git
cd codex-clone

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Accept HuggingFace terms for CodeLlama
huggingface-cli login

Quick Start

Generate Python code

python codex_cli.py --prompt "Write a function to calculate fibonacci numbers" --language python

Generate JavaScript code

python codex_cli.py -p "Create a function to validate email addresses" -l javascript

Save to file

python codex_cli.py -p "Write a REST API endpoint" -l python -o api.py

Usage

Usage: codex_cli.py [OPTIONS]

Options:
  -p, --prompt TEXT              Natural language description (required)
  -l, --language TEXT            Programming language [default: python]
  --length INTEGER               Max tokens to generate [default: 256]
  --temperature FLOAT            Creativity (0.0-1.0) [default: 0.7]
  -o, --output PATH              Save to file
  --help                         Show help

Models

The tool supports multiple open-source models:

1. CodeLlama (Recommended)

   • Better code quality
   • Meta/Facebook trained
   • Requires HuggingFace login

2. StarCoder

   • No login required
   • Good performance
   • Faster inference on CPU

Switch models by editing MODEL_NAME in codex_cli.py:

MODEL_NAME = "bigcode/starcoder"  # Use StarCoder instead

Performance Tips

• Use GPU: CUDA_VISIBLE_DEVICES=0 python codex_cli.py ...
• Lower temperature for consistent code: --temperature 0.3
• Increase length for complex code: --length 512

Limitations

• Generated code should be reviewed before use
• Larger models (13B, 34B) require more VRAM
• First run downloads ~7-15GB of model weights

Contributing

Contributions welcome! Areas for improvement:

• Support for more models
• Fine-tuning for specific languages
• Web UI
• Streaming output

License

MIT License - See LICENSE file

Disclaimer

This tool generates code suggestions. Always review generated code for correctness and security before deploying to production.

Resources

• CodeLlama Documentation
• StarCoder
• Transformers Library