Evidence Conflict Discovery — not "verify the Agent". You stay in control of the decision.
1
Connect Agent
Paste or upload a run record, or integrate the SDK.
2
Collect evidence
VeraNet ingests the claim and available sources.
3
Compare claims vs records
Same event, multiple sources — we map the relationship.
4
Show evidence differences
Where sources disagree, we surface it plainly.
5
Generate conflict report
A reconciliation report for human review.
Live evidence network
Real conflicts, surfaced in real time
Pulled live from /v1/evidence/conflict-graph/query — not a static sample.
…evidence relationships in the live graph
Loading live conflict network…
External submissions are user-reported (external-submission), not independently verified. Independent Witness coverage 0% until L1. We describe relationships; we do not certify them.
We say “Evidence mismatch detected” — never “Agent failed”. No yes/no truth judgment. A conflict candidate becomes a confirmed conflict only after human reviewer sign-off.
The capability
Four states. Not a score.
VeraNet doesn't decide who is right. It describes the relationship between a claim and its evidence — and leaves the decision to humans.
matched
claim & evidence agree
missing
claim, no evidence
conflict
evidence contradicts
unresolved
not yet determinable
VeraNet is not a verifier of agents. VeraNet is the evidence conflict resolution layer for agent evidence.
VeraNet 不判断智能体是否可信,而协调不同来源证据之间的事实一致性。
Building an agent? Bring us a real conflict.
We're looking for 5 founding partners with real agents and real evidence — not 500. Join Early Access and help define the category.