WebGPU Cluster Turns Browsers Into AI Inference Nodes

Alexsandro Souza’s WebGPU Cluster turns browser tabs into shared AI inference nodes, with WebGPU hosts running RF-DETR or SmolVLM behind a simple HTTP API.

0:00
/0:24

A WebGPU Cluster page showing browser GPU sharing, a client-to-broker-to-browser-host flow, and model cards for RF-DETR and SmolVLM.

WebGPU Cluster turns a browser tab into an AI inference host. Pick a model, keep the tab open, and a Node broker can send it detection or image-description jobs through a simple HTTP API.

The current build supports RF-DETR Medium for object detection and SmolVLM-500M for image description, both loaded from Hugging Face and run in the host browser through WebGPU.

It is less about a polished consumer app and more about a useful question: how much shared compute can the web expose when WebGPU becomes the worker, not just the renderer?

Launch Demo
ShowcaseWebGPUAI/MLCompute ShadersWASM