🧠 Quilt Substrate — WebNN API

WebNN (Web Neural Network API) — direct access to neural network ops (matmul, conv, etc.) using hardware acceleration when available. Falls back to WebGPU, then WASM, then CPU.

Backend Detection

Detecting...

MatMul (WebNN Graph Builder)

Build a graph: y = softmax(A @ B). Hardware-accelerated via ML accelerator (CPU GPU NPU).

× ×

Click to test...

MLP Forward Pass

Build a 4-layer MLP graph (Linear → ReLU → Linear → ReLU → Linear → Softmax).

Click to test...

Notes on WebNN support

WebNN is currently:
- Chrome/Edge 112+: behind flag (chrome://flags/#enable-webnn)
- Firefox: behind pref (about:config: dom.webnn.enabled)
- Safari: not yet implemented

Even without WebNN, this page falls back gracefully:
- WebGPU (compute shaders)
- WASM (with optional WASM-SIMD)
- Pure CPU

The substrate primitives work regardless of which backend. The substrate is backend-agnostic.