Many minds need
one set of rules.

The best AI isn’t one model. It’s a few of them. One writes better code, another researches better, another sees design better. So the people getting the most out of AI already use several, and spend their days as the glue between them: copying context, pasting the same brief, re-explaining decisions, and checking whether any of them actually did what they said.

At the same time, AI is learning to act. It can send the email, push the code, book the flight. That’s where things go wrong. Models drift off the task. They report work they didn’t do. They can be talked into things by a web page or an email. Most agent products answer this with a better prompt. A prompt is a request. It isn’t a rule.

What we believe

• Models think. Something else should govern. The rules belong in an engine that isn’t a model and can’t be persuaded.
• Done means proven. A test, a source or a reviewer, not a confident sentence.
• You stay in charge. Anything that sends, buys, publishes or deletes waits for your yes, and your yes covers exactly what you saw.
• Work should outlast the chat. Plans, decisions and progress persist, so nothing starts from zero.
• Your AI should be yours. Any model, including ones on your own desk. Your context is portable, never held hostage.

So we built Ghloo

One conductor that knows you, a squad of the best minds each with its own tools, and an engine underneath that keeps them honest and keeps the work going. It’s free on your own computer, and the parts that decide what an agent may do will be open source.

We’re starting with a private beta, inviting people in waves. If this is how you want to work with AI, we’d love to have you.