Root Who owns the machines
The frustration is earned. The answer is ownership.
People are right to be fed up with big tech. For years, the apps we rely on have watched what we do, kept it, and used it to steer what we see and buy. AI can make that watching sharper. The way out is to own the machines ourselves, so the people they serve are the people who run them.
↓ Scroll
The problem isn’t the technology. It’s who owns it.
Right now, we rent
Nearly all the computing behind modern AI sits in the data centers of a handful of companies. Everyone else rents time on it, on whatever terms they set. When you rent, the owner makes the rules.
Root Compute holds its computing in common instead. It’s a co-op: the members own it, and each member gets one vote, whether they lend a top-end card, an old one or none at all. (While the co-op is being built, the founders hold extra say for a limited time; How it works spells out when it ends.)
The calls a company would make behind closed doors, members make in the open.
What owning it means for you
- Own your own work.Your contributions build something you’re part owner of, not someone else’s profit.
- Your data stays yours.Your prompts and files aren’t logged, kept or sold.
- It keeps them honest.Even if you never join, a member-owned option means the big companies can’t set every term.
What about the other members?
If my request runs on someone else’s graphics card, couldn’t that person see it?
Here is what a host’s machine actually gets:
- No name. Requests pass through a relay that hides who sent them. A host sees a stranger’s words with nothing attached.
- Never the whole story. Requests are spread across many hosts, so no one host becomes the place where your requests collect.
- Any one of them could be a test. The co-op sends decoy requests, and they look like everything else, so a curious host’s question isn’t “what could I find?” It’s “is this one a trap?” Break our privacy rules and you’re out for good. We don’t do second chances. But we do hear evidence. If you believe we got it wrong, you can appeal and show us, and a panel that had no part in the original decision will review it.
Said plainly No system is perfect. This is the plan for launch. Requests marked private go only to machines the co-op trusts most, and the app tells you which route you’re on. Sealed hardware, where even the machine’s owner can’t look inside, is the goal once the co-op can afford it.
Where open models come from
Root Compute runs open models only, like Qwen, Llama and DeepSeek, so anyone can download and inspect them. We won’t pretend they came from nowhere: like nearly every AI model, they were trained on huge amounts of text gathered from the web, and most of the people who wrote it were never asked.
What we can promise is what the co-op does itself.
Root Compute will not scrape anything, and the content of your requests will never be kept.