A neutral second opinion before every irreversible action — and a signed, independently checkable proof after. With a public track record of being right.
Why a third party?
Even frontier-lab guidance now says a maker can't grade its own homework — a separate verifier outperforms self-critique. But a verifier you spawn still shares your incentives. The hardest acceptance test is one with its own public track record on the line.
Ours is live: the verdict ledger — every judgment signed, and relay- and Bitcoin-anchored (committed to Bitcoin proof-of-work via OpenTimestamps) before its outcome settles on-chain. Wins and losses, all public. It is the production reference implementation for ERC-8299 (input provenance for AI inference), which we co-author.
And we hold ourselves to the same bar: we run the open conformance registry where agent verifiers are graded against one recomputable standard, ours included. No green by assertion — a row turns green only when the registry recomputes it from public bytes. The referee is graded too.
What a plan gets you
Per-call governance for your agents: /review verdicts before they act,
signed proofs after, and an audit trail a third party can verify without trusting you —
or us: a recomputable, post-hoc-verifiable record of the kind logging rules like the EU AI
Act (Art. 12) call for. Built for whoever is on the hook for an agent's mistakes.
Not ready to subscribe?
Join the waitlist — tell us what your agents do and what you're on the hook for, and we'll tell you honestly whether this fits.
Pay per call in Lightning, USDC (x402 on Base), or card. Docs: /tool · Proofs: /verify-proof (free, no auth)