Is your agent safe? Prove it.

Cronch runs agents as harnesses whose capabilities have contracts. A rule, like "nothing is sent to a server you haven't allowed", is proved over those contracts, and each proof is checked by a proof checker, not by Cronch. The model is never trusted: the proof holds whatever it chooses.

Run an agent on your machine, with no account: npx @cronch/kernel run. Its source is on GitHub.

Join the waitlist

We're letting people in a few at a time. Already in? Sign in.

Loading sign-in...