An AI approved a refund, moved money, or shipped code. Months later someone asks what happened — and the only record is a log line you could edit. VeriTrace gives every AI decision a receipt that cannot be quietly changed, and anyone can check it in seconds, without an account and without trusting us.
No signup, no email. Both open in the browser and re-check themselves while you watch.
It approves a refund, blocks an account, or deploys code. Today the only record is a log line — and whoever controls the log can change it later, with no trace.
Each decision gets a receipt holding a fingerprint of what went in and what came out, a timestamp, the exact model version — then a cryptographic wax seal over the whole thing.
No account, no API key, no phone call to us. Change one character anywhere in the receipt and the seal breaks — visibly, for everyone, forever.
One decision from a demo agent. Signature, batch membership and the judge’s verdict all check out.
Inspect it →✗ TamperedSame kind of receipt, but the output was edited after signing. Watch the check fail and tell you exactly which layer caught it.
Watch it fail →Both receipts are produced by the real open-source SDK — signed, Merkle-batched and verdict-signed for real. The underlying decisions are demo data, not customer traffic. The cryptography is genuine; the business is a demonstration.
How often each agent’s decisions were judged correct. Only receipts that still pass verification are counted, so a doctored receipt cannot improve its own score.
Green means the signature still matches. Click any row to see the receipt itself and every check it went through — including the deliberately altered one, whose signature no longer matches. Scores are computed live from verified receipts only, and a tampered receipt’s self-reported verdict is never trusted.