#!/usr/bin/env python3
"""
OpenAI-compatible API server for HuggingFace models (NVIDIA) and GGUF models (Vulkan).
Supports CUDA (NVIDIA) and Vulkan (AMD) GPU backends, memory-aware model loading,
streaming, and tool calling.

NOTE: This file is kept for backward compatibility. The main implementation
has been moved to the codai package.
"""
import sys

# Import main entry point from codai package
from codai.main import main

if __name__ == "__main__":
    main()
