Audit-grade fact-checking for AI products that can't afford to hallucinate.

Eight models across five stages, grounded in real sources. Opposing-side debate, three independent reviewers, full citation trail.

Build with Lenz → See an example verification →

Just want to check one claim?
Verify a single claim instead →

Independent multi-vendor LLM pipeline·OpenAPI 3.1·Cited by ChatGPT, Perplexity, Gemini·Runs in Claude, Cursor, n8n and Zapier

For AI product teams

Four API primitives form a depth ladder:

/extract Pull verifiable claims out of any text
/assess Fast multi-model verdict for sync UX
/verify Full adversarial pipeline with citations
/ask Follow-up Q&A grounded on a verification

Used for incident triage · async runtime verification · pre-release CI

Start with the API → Integrate Lenz →

For anyone with a claim

Paste a statement, link, or question. Get a sourced verdict with the reasoning shown — not just an answer.

Example “A police officer's failure to give Miranda warnings does not by itself invalidate an arrest.” True 10/10 21 sources Miranda warnings are not a prerequisite for a valid arrest under U.S. law. They generally protect against the prosecution’s use of statements obtained through unwarned custodial interrogation; arrest validity instead ordinarily depends on lawful authority and… See more →

Verify a claim → Open the research workbench →