falcon_account
ChatGPTclaims calls no model and is never counted.By One Intelligence LLC
FALCON Verify checks whether an AI's answer is supported by the evidence it was written from, and tells you claim by claim. Ask Claude to verify what it just wrote against the document, figures or records you gave it, or paste an answer and its source. FALCON extracts the material claims, settles what it can deterministically (numbers, names and dates against the evidence) and sends the rest to a judge model. Every claim comes back with a verdict: supported, qualified, unsupported, contradicted, or more evidence needed, with the reason. It verifies against the evidence you supply, never against the model's own knowledge, so a confident answer with no support is reported as exactly that. Every result carries a hash-chained record of the decisions that produced it, which anyone can check. Typical uses: checking a generated summary against its source before sending it; confirming a report's figures match the records; reviewing a draft against the policy or contract it describes. Free for 25 verifications a month, with Starter, Personal and Team plans after that. Nothing to install: sign in with Google, connect, and ask.
claims calls no model and is never counted.falcon_verify needs a model that can follow an extraction format, and on a substrate without one it refuses rather than returning an empty report — call this first and you learn that before you depend on it. The model check is a real probe, not an inference from the driver's name: a substrate with no API key configured looks identical to a working one until it is asked.insufficiency block answering what was established, what was not, why, what evidence or depth would resolve each one, and what to do next. and its span in the source text. Answers 'is this supported by the evidence you provided' -- NOT 'is this true': a claim the evidence cannot speak to comes back UNVERIFIABLE, never supported. The unsupported rate is over claims that were decided, and is null when none were, because a response nothing could be checked against is not a clean one. Every call leaves a hash-chained audit record, returned as audit, which pins the digests of what was verified without retaining the text, and an attribution line naming FALCON Verify, the run id and the counts, with report_as saying how a report cites it. Needs no domain configuration.run_id that falcon_verify returned under audit. state is one of queued, judging, settled, partial or expired. unsupported_rate stays null until the run reaches a terminal state, with provisional_rate carrying the rate over whatever has been decided so far; polling until unsupported_rate is non-null is the intended pattern. A partial run stopped early -- usually at its cost cap -- and reports an interval, because the claims it judged are a uniform sample of the output rather than its first N.Add this badge to your site or README. It links to this page.
HTML
<a href="https://mcp-app-store.keenethics-labs.com/app/falcon-verify"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>
Markdown
[](https://mcp-app-store.keenethics-labs.com/app/falcon-verify)