Back to Devoured - August 10, 2026
DEVOURED
jax-js (Website)

jax-js (Website)

AI webjavascript Jax-js.com
jax-js brings JAX-inspired numerical computing and machine learning to the browser with WebGPU and WebAssembly support.
What: jax-js is a zero-dependency JavaScript library that enables client-side execution of neural networks and numerical algorithms. It features a lightweight compiler to generate WebGPU and WebAssembly kernels, supporting local inference for models like Whisper and D-FINE.
Why it matters: This indicates a growing trend toward running computationally intensive AI tasks locally on consumer devices, bypassing cloud latency and hosting costs for specific use cases.
Takeaway: Run `npm install @jax-js/jax` to begin prototyping browser-based ML models.
Deep dive
  • Performance: Achieves high throughput via WebGPU and WebAssembly compilation.
  • Portability: Runs in all major desktop and mobile browsers.
  • API: Modeled after JAX and NumPy, simplifying the transition for data scientists.
  • Capabilities: Supports LLM chat, image processing, fluid simulations, and training.
  • Deployment: Allows for end-to-end client-side ML without external API dependencies.
Decoder
  • WebGPU: A modern web standard providing high-performance access to the underlying GPU for graphics and general-purpose computation.
  • WebAssembly (Wasm): A binary instruction format that allows code written in languages like C++ or Rust to run in web browsers at near-native speed.
  • JIT (Just-In-Time) compilation: A method of executing code by compiling it to machine instructions during execution rather than prior to it.
  • Kernel: A small program designed to run on a processor (CPU or GPU) to perform a specific, high-performance mathematical task.
Original article

jax-js is a machine learningan ML library and compiler for the web

High-performance WebGPU and WebAssembly kernels in JavaScript. Run neural networks, image algorithms, simulations, and numerical code, all JIT compiled in your browser.

Add jax-js to your project

Zero dependencies. All major browsers, with bundlers and in JS modules.

npm install @jax-js/jax

Matrix multiplication

Billions of floating-point operations (GFLOPs) per second

Like JAX and PyTorch in your browser

jax-js is a end-to-end ML library inspired by JAX, but in pure JavaScript:

  • Runs completely client-side (Chrome, Firefox, iOS, Android).
  • Has close API compatibility with NumPy/JAX.
  • Is written from scratch, with zero external dependencies.

jax-js is likely the most portable GPU ML framework, since it runs anywhere a browser can run. It's also simple but optimized, including a lightweight compiler that translates your high-level operations into WebGPU and WebAssembly kernels.

The goal of jax-js is to make numerical code accessible and deployable to everyone, so compute-intensive apps can run fast and locally on consumer hardware.

Live examples

Interactive demos for local AI models, simulations, geometry processing, and numerical methods.

LLM Chat

Run a small language model locally in your browser, with GPU and CPU-based inference.

Neural Cellular Automata

Grow image patterns with a tiny differentiable automaton.

Heat Method Geodesics

Compute surface distances on triangle meshes with dense solves.

Whisper ASR

Transcribe audio locally with OpenAI Whisper.

Kyutai Pocket TTS

Voice cloning AI model that runs in your browser.

Fluid Simulation

Interactive Navier-Stokes flow with WebGPU kernels.

D-FINE Detection

Run an ONNX object detector locally with WebGPU.

MobileCLIP2 Inference

Compute embeddings for book passages.

MNIST Training

Train a neural network on handwritten digits in-browser.

Principal Component Analysis

Explore components of a 3D point cloud.

Benchmarks

Compare jax-js kernels with other web ML libraries.

Resources

GitHub Repository

Check out the code and tutorial.

REPL

Try jax-js in this browser-based REPL.

API Reference

View the generated API documentation.

Explore all 68 articles in today's digest

Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi