Stop AI agents from taking risky actions — before they happen.
Runtime authorization: your contract checks a signed risk verdict before an agent acts, without trusting the caller.
treasury transfer
APPROVED · trust 82
audit ✓ · quorum ✓ · nonce 12
ecrecover ✓
~3k gas · replay-proof
Proceeds
your policy allows
You decide
never a silent pass
Authorization flow — how a verdict gates an action
Agents act on real assets. Trust is assumed.
Contracts governing autonomous agents usually can't check whether the code was audited, the multisig is sane, or the data is trustworthy.
A verdict the contract verifies itself.
Every input proven at source, cross-checked across providers, scored, signed — and recovered on-chain with ecrecover. Unprovable inputs return UNKNOWN.
Proven inputs → your enforcement.
Policy, verdict, contract call, failure modes.
Only provable inputs are scored
Domain-separated, replay-proof
Your contract verifies, not us
Fails closed, never silent
Verified against a real EVM
Advisory, by construction
The same three views, on every product.
An evidence card, a timeline and a trust score — identical components everywhere, so what you learn on one product transfers to the next.
What changes after you adopt it.
Agent actions gated on proofPermissions rest on audited code and cross-checked state — checkable by anyone.
Replay attacks off the tableDomain separation, nonces and staleness windows, verified against a real EVM.
Community-verifiable trustYour users verify the same verdict your contract did.
Governance keeps the wheelAdvisory by design: pause logic and overrides stay in your contracts.
Zero honor system, end to end.
Nothing caller-supplied
Inputs without an un-spoofable source are removed, not estimated.
Multi-provider quorum
Provider disagreement or stale feeds → UNKNOWN, never a guess.
Replay-proof
Chain id + contract bound in the domain; monotonic nonce; staleness enforced.
Real-EVM verified
solc 0.8.24 · full on-chain test matrix passes.
Composable
isAgentActive() is a view your protocol consults — authority stays yours.
Freshness on-chain
An APPROVED past max age reads inactive automatically.
Axiom is advisory — deliberately not a kill-switch. It supplies the verified verdict; execution authority stays with the consuming protocol's governance. Axiom itself never takes an on-chain action.