Free · no signup to try

Is your AI making things up?

Paste an answer your AI gave a user. We pull out every factual claim and check each one independently against the live web — then flag the hallucinations, with the receipts. Built for teams shipping AI in law, health, insurance, and finance, where a made-up fact is a liability, not a typo.

Checks up to 8 claims · usually 1–2 minutes · we don't store your text
The fake case in it, Varghese v. China Southern, is the one that got two real lawyers sanctioned.
Also free · no signup

Do your own two pages disagree?

Give me two URLs from the same company and I'll tell you where they contradict each other. Not whether either is right — I can't know that, and neither can anyone else from outside. Only that both can't be. That part is checkable: every finding comes with the exact sentence from each page, and a command you can run yourself.

A pricing page and a docs page is the usual pair · 3 checks an hour
Two pages from a real company that disagreed until someone told them.
Free · no wallet · for x402 sellers

Can an agent actually pay your endpoint?

Paste a paid URL. I'll read its 402 the way a spec-current buyer would and tell you whether one could construct a payment from what came back, and exactly what's missing if not. We measured 1,521 x402 hosts to build this: 92 return their challenge in the body with no PAYMENT-REQUIRED header, so a v2 buyer reads nothing and leaves — while v1 buyers keep paying, so nothing in your own logs ever says why the rest stopped. We shipped that class of bug ourselves and only caught it by checking. No model is involved here, so every line of the answer is reproducible with the curl command it hands you.

One unpaid probe · nothing is spent, yours or ours · 10 an hour
Our own endpoint, including the manifest check that caught our bug.

Don't take our word for it. That's the product.

Anyone can tell you an answer is wrong. The question is why you'd believe the checker. Every paid verdict ships an Ed25519-signed receipt that binds the exact claim, the verdict, and the evidence behind it. You can verify that signature offline, against our published key, without an account and without ever calling us again — check one here, or run verify_receipt.py from the open-source repo and don't trust our server at all.

A signature also isn't permission. require_receipt refuses a receipt that is genuine but issued over a different claim, or stale, or already spent — so an agent can't claim it checked something it didn't.

This free scan is unsigned. It runs the same checks against the same live web; the receipt is what a paid call adds.