A universal, vendor-agnostic AI inference runtime. Run any model on any hardware with the .nbx universal format. Zero fallback, blazing fast.