Universal LLM deployment by CMU + Apache TVM. Compiles models to run natively on iPhone, Android, WebGPU, Vulkan, Metal, and AMD — anywhere, fully offline.