dehallucinator extracts every claim, statistic, date, and quote from a document and verifies each one against the open web — returning an evidence-backed, per-claim report. Engineered to abstain rather than wrongly accuse.
Exact numeric, quote, and date checks constrain the LLM judge — a claim is only refuted on positive contradiction from independent authoritative sources.
Every verdict cites its sources, shows confidence, and proposes a concrete correction. Six labels keep distinct failure modes from collapsing into one score.
Watch verdicts land live in the dashboard, or call the metered REST API with a key. Usage is tracked per month against your plan.
Every claim, statistic, date, quote, and citation is pulled out and rewritten to stand on its own.
Hard facts — dates, country statistics, citations — are checked against authoritative databases (Wikidata, the World Bank, Crossref) deterministically, in milliseconds and with no model. Everything else is verified against the live web.
Each claim gets an evidence-backed verdict and the document a trust grade — and we abstain rather than wrongly accuse.
Start free. Scale when you need more verified claims.
Paid plans are display-only for now — every account runs on the Free tier until billing is enabled.