Waffuru

ワッフル

Low-level tools for running models fast on the hardware you already own.

Coming soon

Iron

A Rust kernel DSL. Press raw math directly into high-performance silicon: one definition compiles down to Metal, CUDA, HIP and Vulkan.

Coming soon

Butter

Melt away latency with ultra-smooth local inference. A dependency-light LLM inference library for Apple Silicon, built on pre-compiled Metal kernels from Iron. No Python, no MLX, no JIT.

Releasing shortly · github.com/waffuruai