WebNN API polyfill

React Native WebNN

Lenout Renderer (night) — DPI-aware tile-based WebGPU renderer with WebNN refinement and WebGL2/Canvas fallbacks.

Headless client-side AI video-generation engine for the browser. Runs LCM / SD-Turbo diffusion on WebGPU or WebNN (onnxruntime-web), CLIP tokenization via @huggingface/transformers, Mamba SSM temporal coherence, and WebCodecs MP4 muxing. No React. For a r

WebNN provider for browser-ai — hardware-accelerated inference

TypeScript types for WebNN

A tflite delegate based on Web Neural Network API (WebNN API).

Training and Evaluation AI Model Recommendation Engine for WebNN

LiteRT.js package

Cross-platform NPU (Neural Processing Unit) detection for JavaScript — Node.js + Browser (WebNN)

GPU Powered Neural Networks for every Browser

Convert ML models (.onnx, .tflite) into ready-to-run WebNN API JavaScript/TypeScript code

The jQuery of local AI — A unified API that abstracts browser AI runtime fragmentation (Native / WebNN / WebGPU / WASM)

Full embeddable AI Video Studio for React — <StudioPanel> drop-in that runs LCM / SD-Turbo diffusion on WebGPU/WebNN, conditioned on a Mamba SSM state for temporal coherence, and muxes MP4 via WebCodecs. React layer on top of the headless @seanhogg/builde

PyTorch-shaped Python library that traces ops into a UOp IR and realizes through NumPy. The JIT epoch successor to browsergrad-grad — same API surface, lazy by default, designed for downstream fusion / WGSL / WebNN passes. Educational source code, MIT-lic

Test for deliver TFLite WebNN external delegate WASM binary

A Zero-Dependency, Bare-Metal Deep Learning Engine for the Browser and Node.js

Web primitive layer implementations for xPU: one reference per op, one implementation per backend, and tests that hold them together.

YOLO browser inference client powered by onnxruntime-web

Local Inference internal utils package to avoid boilerplate across codebases.