Prove your agent answers
Verify that every claim your agent makes is backed by evidence from its context.
- Boxed warning for thyroid tumours
- Injected once a week
- Most pharmacies keep it in stock
- Halves cardiovascular risk
Powered by 50+ data providers
How it works
Call /verify with the claims and their context. Get a verdict per claim, with the evidence behind it.
Send the claims
Pass the sentences your agent wants to send, one claim per item. The call runs before the answer leaves.
We read the sources
Search routes each claim to the providers that hold the answer: drug labels, registries, filings. Verify compares the source text with the claim.
Get a verdict per claim
Supported, contradicted or unverifiable, each with a quote from its source. A contradicted claim comes with its correction, so your agent sends only what holds up.
Verify, from your terminal
Check a claim against its sources, or verify an agent’s full answer against its context. With the KIRHA skill, your coding agent can run the check before responding.
Works in any harness
Get your API key, then install with one command
- π
Why an endpoint
Find the source, compare, cite. Every team with an agent in production rebuilds that loop. Verify runs it on authoritative data.
Checked against the source
An LLM judge can only say whether an answer sounds right. Verify reads the label, register or filing the claim is about.
Citations that point somewhere real
Every quote is checked against the provider output it cites. A quote that is not there is dropped, and a verdict left without one becomes unverifiable.
Unverifiable is an answer
When no provider covers a claim, you get unverifiable instead of a guess, and your agent can hedge or leave the claim out.
Every check improves coverage
Claims no provider covers show us which providers to add next. The graph grows where agents need it.
Ship answers you can defend
Put Verify between your agent and its users. Every claim that goes out has a source behind it.