Before any agent or machine acts, it asks Cascade one question: may I take this? Cascade works like a referee — one answer, the same every time, in under a millisecond of decision compute — and signs a receipt anyone can check later, even offline. Any model, no extra LLM tokens.
Each of these would make Cascade easier to demo and harder to trust. The same inputs only give the same answer if nothing in the decision path can change.
Your AI or planner decides what should happen. Your agents and machines make it happen. Cascade is the referee in between — the one part that has to be exactly right every single time.
Nothing above it changes. No migration, no new message bus, no rewrite of your autonomy stack. One call before you act.
Every decision gives the same answer for the same inputs and is signed the moment it is made. So the receipt comes from how Cascade works, not from a separate log that could disagree with it. That is why anyone can check it — even offline, without trusting us.
Start free with 1,000,000 lifetime signed decisions, no card. Go to production on Builder for $49 a month, and cancel anytime. Need more? Add top-ups at $10 per 250,000 decisions. They never expire. The meter is the count of signed receipts in your own record, so you can recompute any invoice yourself.
One signed decision per claim. each making decisions an hour, around the clock, is decisions a month.
Every grant and every hold is a decision. Releases and waiting are free. The free beta covers 1,000,000 lifetime signed decisions, no card. Builder covers 2,000,000 a month for $49. Token figures are an estimate: about 600 LLM tokens per decision if a model refereed it, at $3 per million tokens.