WebGPU inference grid

TeamThink

Start a session, invite devices with a link, and run model inference across a peer-to-peer mesh. Each device that joins becomes a compute node; requests are routed to whoever has the GPU to spare.

Host a session

Create a new grid and get an invite link to share.

Join a session

Paste an invite link or enter a session code.

How it works

  1. 01Devices connect directly over WebRTC; a tiny rendezvous only brokers the initial handshake.
  2. 02Each node gossips its capabilities and load; shared session state is replicated across peers as a CRDT.
  3. 03Capable nodes claim open requests, run them on WebGPU, and stream tokens back to the requester.