WebGPU network

How it works

This demo explores a worldwide network of AI models running in people's browsers. A model in one browser can answer a prompt from another.

JOIN
When you open the page, your browser joins through Cloudflare as a leader or consumer. Your device, saved model files, and network demand help decide.
LEADER
Downloads Qwen3 1.7B if needed, saves its files in private browser storage (OPFS), and runs it on its GPU with WebGPU. It answers requests one at a time.
CONSUMER
Sends prompts through Cloudflare to an available leader and receives the answer as it is generated. It does not load the model.

Tabs in the same browser share one connection and avoid duplicate model loads. Cloudflare routes requests; it does not run the model.

THIS BROWSERCONNECTINGChecking coordination...-
THIS BROWSER'S WORKER TABElectingWaiting for Web Lock
MODELWaiting-
NETWORK 0 leaders | 0 consumersWaiting for network role

Prompt

Qwen3 1.7B | thinking enabled | browser inference | Ctrl+Enter or Cmd+Enter to send

Response

No prompt yet.