Jax Js

This library is a machine learning library and compiler for the web, written from scratch in pure JavaScript with zero external dependencies. It is inspired by JAX and offers close API compatibility with NumPy and JAX, and it runs entirely on the client side in browsers including Chrome, Firefox, iOS and Android. A lightweight compiler translates high level array operations into WebGPU and WebAssembly kernels that are compiled just in time in the browser. The vendor positions this as a portable GPU machine learning framework that runs anywhere a browser does, so that neural networks, image algorithms, simulations and numerical code can run locally on consumer hardware. It installs as a single scoped npm package. The site hosts a live editor and a browser REPL, an API reference generated from the source, a benchmark page comparing its kernels with other web machine learning libraries, and a set of interactive demos. These include a small language model chat, Whisper speech transcription, a voice cloning text to speech model, an ONNX object detector, MobileCLIP2 embeddings, and MNIST training in the browser. Web developers who want to ship local inference or numerical computing without a server, and researchers who want to publish runnable examples, are the natural audience. The library is distributed through npm with its code on GitHub, and the vendor publishes no pricing.