**Zinc** is a Zig-based inference engine for local LLM execution on AMD RDNA3/RDNA4 GPUs via Vulkan and Apple Silicon via Metal, producing one binary without ROCm or MLX.

It supports models like Qwen3 8B Q4_K_M and Qwen3.5 35B Q4_K_XL, offering CLI inference, OpenAI-compatible API, and browser chat UI. Build with Zig, manage models via CLI commands, and run checks for GPU/shader verification on Linux or macOS.