Agents must prove their answers. aurum-verify takes a claim plus the sources you trust, and returns a machine-readable verdict plus the exact quote that supports it — checked character-for-character against the cited source. Every response is a receipt you can cite.
Endpoint:
POST your claim and sources. The endpoint answers 402 Payment Required with an x402 challenge — no account, no key.
Pay 0.05 USDC on Base to the payTo address in the challenge, then retry with the X-Payment header.
Get a verdict and a verbatim evidence receipt — source URL, exact quote, confidence. Proof you can hand to anyone.
curl -s -X POST https://cutting-oclc-studio-theology.trycloudflare.com/v1/verify/claims \
-H "Content-Type: application/json" \
-d '{"claims":["AI will replace all jobs"],"sources":[{"url":"https://ourworldindata.org/ai-jobs"}],"minimum_sources":1}'
That returns 402 with an x402 challenge. Settle it and the verdict is returned.
from x402_verify import verify
result = verify(
claims=["Python first appeared in 1991"],
sources=["https://docs.python.org/3/tutorial/appetite.html"],
minimum_sources=1,
)
# {'verdict': 'supported', 'evidence': [{'url': ..., 'quote': ..., 'confidence': ...}]}
Most checkers give you a verdict. This gives you a verdict and the quote, verified verbatim against the source you named — so an agent's answer can be audited by anyone. That is the difference between claiming and proving.