Paste text, drop a link, check any page. Veriq extracts every checkable claim, hunts down real sources — including ones that disagree — and returns a verdict backed by verbatim quotes. No vibes. Only evidence.
"The Eiffel Tower is 330m tall."
3 sources · 94%"Coffee causes dehydration."
5 sources · 61%"The Great Wall is visible from space."
4 sources · 96%Every claim passes through six stages. The system actively tries to disprove each claim before it dares call anything true.
Compound text is broken into atomic, checkable claims. Opinions, predictions, and vague fluff are tagged uncheckable and set aside — never graded.
Each claim is searched for supporting evidence and with disproof queries designed to refute it. A verdict that never met its counter-evidence isn't a verdict.
A small embedding model (bge) scores every candidate source for claim-fit by cosine similarity. Cheap, fast, no LLM needed.
Fit × deterministic credibility — source tier (official, press, blog…), recency. Credibility is computed from observable facts, never judged by an LLM.
Source pages are fetched and the best quotes extracted. Each quote is judged supports / refutes / neutral — and must be a verbatim substring of its page, enforced in code.
Deterministic aggregation into four states: supported · contested · refuted · unverifiable — with confidence, counter-evidence, and a "what would change my mind" line on every receipt.
Every verification includes a red-team pass that hunts for refuting evidence. Confirmation bias is a bug; we test for it.
Each verdict ships the exact quotes, links, and source tiers behind it. Click any citation and read the source yourself.
Binary true/false is a lie. Supported, contested, refuted, or honestly unverifiable — nuance is the product.
Source pages are snapshotted at verification time, so receipts stay reproducible even when the web changes.
Web app, Chrome and Firefox extensions, and a REST API. Verify a page without leaving it.
The whole stack runs on Cloudflare's free tier. Verification shouldn't need a data center — or your credit card.
No exotic crypto, no hand-rolled primitives. Standard building blocks, applied carefully.
PBKDF2-SHA256 with per-user 128-bit salts. Timing-safe comparison on every login.
256-bit random tokens; only hashes stored server-side. HttpOnly, Secure cookies, 30-day expiry.
Per-IP limits on auth endpoints, per-user daily verification caps. Abuse gets boring fast.
Your verifications are never used to train models. Your data is yours.
Sessions expire and die. Rate-limit counters are ephemeral. Nothing lingers.
The pipeline, scoring, and verdict logic are MIT-licensed. Don't trust us — read the code.
An AI answering from memory can hallucinate with total confidence. Veriq retrieves live sources, extracts verbatim quotes, and shows you the evidence — including evidence that disagrees. You grade the receipts, not the vibes.
It means no checkable evidence was found — not that the claim is false. It's a first-class verdict, because "I don't know" is the only honest answer when the evidence isn't there.
Source credibility is tiered deterministically — official documents, papers, and established press outrank blogs and unknown sites. A verdict needs agreement across credible sources, and every citation is click-to-verify.
No. Your verifications are stored only to show your history and serve shareable receipts. Nothing is used to train models.
The entire stack runs on Cloudflare's free tier, so the marginal cost of a verification is near zero. Free accounts get 50 verifications a day — plenty for real use.
Free forever plan. No credit card. Your first receipt is a minute away.
Create your free account