What Peer-to-Peer Browser Compute Is

Peer-to-peer browser compute is a way to pool several ordinary devices into one machine, using nothing but the web browser each of them already runs. No install, no account, no data center in the middle. If you have ever been on a video call, you have already used most of the machinery underneath it.

How it works

The devices connect directly to each other. A small coordinating server does one job, introducing the browsers so they can open a direct channel, and then it steps out of the way. It never handles your data. Once the circle is formed, the work is split across it. For an AI model, that means a model too large for any single device is cut into stages, each device holds some of the layers, and only a small vector of hidden state crosses the wire from one device to the next as each word is produced.

The coordinating server introduces the devices and then leaves. The work, and the data, stay inside the circle.

Why it is possible now

Two things that used to require native, installed software now run inside a plain browser tab. The first is direct device to device connections, the peer-to-peer channels that carry the data between machines without routing it through a central server. The second is real computation on the graphics chip, which lets a browser run the heavy math a neural network needs at usable speed. When those two arrived in the browser, pooling a circle of everyday devices stopped being a research demo and became something you can open with a link.

What it is good for, and what it is not

It is good for pooling capacity you already own into something that can carry a real workload, and for keeping private work private, because on a local network the data never has to leave the building. It is not a way to beat a data center on raw speed for a single request. Spreading a model across devices trades some speed for the ability to run something no single device could hold, and the cost of that trade is latency. We measured exactly what it costs, honestly, in the numbers behind Osiris Compute.

Osiris Compute is our implementation of this idea. The public grid is a free utility you can open in a browser, and the same engine runs as a private grid on hardware a business controls. It is open source, so you do not have to take our word for any of it.

See the grid run, or read the code.