The first product built on Fidenta
Your assistant sounds exactly as confident when it is wrong.
That is not a flaw you can prompt your way out of. A model checking its own work brings the same blind spots to the check that it brought to the answer. Fidenta Verify connects to Claude or Perplexity and checks specific claims against live sources — then has a model from a different lab audit that judgement before you see it.
What comes back
Most fact-checkers collapse “we found nothing” into “false.” That is a lie in both directions — it brands unresearched claims as wrong, and it launders absence of evidence into disproof. We keep them apart.
The evidence affirms the claim as stated, with the sources it came from.
The evidence affirms something incompatible with the claim.
We searched and found nothing either way. Not the same as false, and never reported as such.
An opinion, a prediction, or a matter of definition. We label it rather than pretend to rule on it.
Every verdict carries its sources, and where it matters a caveat — the technically-true-but-misleading part a confident answer glosses over. Ask whether a platform requires a certified device and the quick answer is yes; the useful answer is that it is a support-policy requirement, not a technical one.
How it works
Not a committee of models voting. Only the retrievers have eyes — two models arguing about a claim with no evidence in front of them produce two opinions and a coin flip, and their agreement measures shared training data rather than truth.
Results are remembered on your account, so a claim you have already checked comes back instantly — and if the answer changes later, you are told it changed.
A real run
We pointed it at an AI-written account of the Battle of Belleau Wood — a well-documented event, described by a capable model, with nothing obviously wrong in it. It returned twenty-four claims: fifteen held, eight were contradicted, and one could not be confirmed either way. None of the eight looked like errors.
“Come on, you sons of bitches. Do you want to live forever?” — attributed to Sergeant Dan Daly, repeated in Marine Corps lore for a century.
Contradicted. The line traces to war correspondent Floyd Gibbons in 1918, who attributed it to an unnamed gunnery sergeant. The Marine Corps History Division itself labels the Daly attribution a legend. Daly’s two Medals of Honor, which the same sentence asserted, checked out fine.
German casualties “estimated at more than 10,000.” Wikipedia says it. Britannica says it. It is in every general reference.
Supported — at 65% confidence, with a caveat. The estimate is genuinely widely cited, so the claim as written stands. But archival divisional records put the figure closer to 3,518. The claim is not false. It is repeating a number that better sources do not support — which is a different problem, and the one that actually gets you in trouble.
That German troops at Belleau Wood called the Marines Teufel Hunden — “Devil Dogs” — and that this is where the nickname comes from.
Contradicted. Marine Corps historical research finds no evidence that German troops coined or used the term at Belleau Wood — and the phrase turns up in American newspapers before the battle, in an April 1918 wire report. A century of institutional repetition does not make it a source.
“Ten Medals of Honor were awarded, nine to Marines and one to a Navy hospital corpsman.”
Contradicted. The count was right. The split was right. The recipient was a commissioned Medical Corps officer, not an enlisted corpsman. A reader skimming for plausibility would never catch it.
None of these were hallucinations. They were repetitions — things the model absorbed because the whole internet says them. That is exactly the failure a model cannot catch in itself, because the same training data produced both the answer and the confidence.
Now substitute the market-size figure in your next proposal, or the regulation you are about to tell a client applies to them.
Setup
Fidenta Verify is a connector. You add one URL to the assistant you already use, approve the connection, then ask it to verify something.
https://verify.noevant.ai/mcpWorks on the free plan, which allows one connector.
https://verify.noevant.ai/mcp, transport Streamable HTTPRequires a paid Perplexity plan.
Full setup notes, the tool reference and answers to the usual questions are in the documentation.
Other assistants are coming. Copilot and Gemini currently expose connectors only through an administrator console rather than in the chat window itself, so we are not claiming they work yet.
Early access
40 to start
Then 10 claim checks every month, so the connector stays useful even when you are not paying. No card, no expiry, no trial clock.
$19
100 additional claim checks. Credits do not expire and are used only after your monthly allowance runs out.
One credit checks one claim. A single sentence costs one; paste a news article and it might find fifteen. You are told the number and the price before anything is charged, and you can cap it — “just check the first five” — on anything long. Nothing that fails to come back is billed.
One verification covers up to two claims. We are in early access and setting long-term pricing from real usage rather than a guess — if that changes, existing accounts hear it from us first.
Your account
Sign in at verify.noevant.ai/login with your API token, or ask us to email you a six-digit code if you have mislaid it. From there you can see how many verifications you have left, add more, replace your API token if it leaks, and pull up your receipts.
Top-ups are one-off payments. There is no subscription, no card kept on file until you buy something, and nothing to cancel.
Who it is for
Consultants sending a client a recommendation. Solo operators writing something with their name on it. Anyone who has pasted an AI answer into a document and felt a flicker of doubt they did not have time to chase down.
It is not a content filter and not a plagiarism checker. It answers one question: can this specific claim be supported right now, by a source you can open?
40 free claim checks to start, then 10 a month. No card. Two minutes to connect.
Start free