Run your favorite LLMs (Qwen, Gemma4, NemotronH, DharaAR, ...) at blazing-fast speeds on Apple Silicon, with multi-modal (image/audio/video) support, from JavaScript/TypeScript.

Route your test suite through every AI provider hosting your model; compare cost-per-passing-outcome and trajectory quality locally.

Node.js bindings for llama.cpp with QBNN architecture support

Check what a quantized tensor actually means: GRIT descriptors carry the full numeric contract (element format, group sizes, scale formats, zero points, sparsity, rounding) as 64 bytes with a 64-bit id, verifiable against the bytes in O(1). Spec v1, zero