Give an autonomous agent a spending policy, with your own numbers — then watch it allow and refuse the actions you throw at it, record a decision on the live chain, and catch yourself trying to fake it. Every call below is real and runs against the live network.
Grant the agent a budget and an expiry. It may spend up to your ceiling, until it runs out or the clock runs out — enforced on every action.
Enter an amount and try the action. Under budget → allowed. Over what's left → refused in real time, no human in the loop. Fire as many as you like and watch the meter above.
Write whatever the agent decided. Cinchor commits a fingerprint of it to the chain — the full text stays with you; only the hash is anchored, so it's provable but private.
First verify the record as written: the recomputed fingerprint matches the chain. Then change anything below and verify again — the faked version produces a different fingerprint and fails. That's the whole point: no one, us included, can rewrite it.
Now tamper with your own record and re-check: