Why We Built Osiris Compute
AI is getting more powerful every month, and more centralized every month. Those two trends are not a coincidence. The biggest models live in a handful of data centers owned by a handful of companies, and you reach them by paying per token through an API. It is rented capability, and the meter never stops.
We think that is the wrong shape for a tool this important. So we built Osiris Compute, and today it is open source.
The idea
Osiris Compute pools a circle of ordinary devices, the ones you and the people you trust already own, into one machine that can carry a real workload. It runs in the browser. No install, no account, no crypto, no cloud bill. A model too big for one phone gets split across several, and only a few kilobytes of hidden state cross the wire per token. The coordinating server introduces the devices to each other and then steps out of the way. It never sees your data.
Why we are doing it
Two reasons, one practical and one we believe in.
The practical one is physics. In a distributed model the only thing crossing the network each token is a small vector, so you are never bandwidth bound. Put the circle on one local network and the round trips nearly vanish, which is the same condition a data center runs in. Fast and private turn out to be the same problem, and we wrote a whole piece on that: your data should never leave the building.
The one we believe in is simpler. Capability you rent can be taken away. Capability you own is yours. We would rather hand the tools to the people doing the work than meter them forever. That is the same instinct behind how we think about AI in general: use it to sharpen people, not to replace the thing that made them good.
What you can do with it
For anyone, the live grid is a free public utility. Open a circle, share a link, and the browsers do the rest. For a business, the same engine runs as a private grid on hardware you control, which gives you sovereign AI for the everyday work and a way to pool idle desktops for heavy jobs, without your data ever leaving the premises.
The engine is AGPL and on GitHub. Fork it, host it, federate it.