COMING SOON

[Michael-A-Kuykendall]

[Michael-A-Kuykendall]

LONG DELAYED but coming soooooooon

Shimmy v2.0 ships with Airframe — a from-scratch GPU inference engine written in pure Rust with no llama.cpp dependency, no C FFI, no Python, no CUDA.

How it works:
-Loads GGUF models directly (custom reader — no third-party parser)
-Runs transformer inference on your GPU via wgpu (D3D12 on Windows, Metal on macOS, Vulkan on Linux)
-Ships 7 hand-written WGSL compute shaders — one per pipeline stage
-Exposes OpenAI-compatible, Ollama-compatible, and Anthropic-compatible APIs