Grounded mode
Real entailment inference over (your context, the claim). A solved, scoreable problem: does the claim actually follow from what you supplied?
A pay-per-call endpoint an agent can hit mid-loop: "verify this claim before I act on it or hand it downstream." Every verdict comes from a real model inference — or, in open mode, a real live search plus real inference over the results — run at request time. Never cached, never fabricated.
"Open" mode checks consistency with live web evidence found right now — not objective truth. A claim can be well-supported by current web content and still be wrong. Read the full scope in the README before treating a verdict as more than that.
A small, isolated, swappable scoring module — not a black box. Grounded mode runs HHEM-2.1-Open (Apache-2.0, CPU inference) directly; open mode adds a real Tavily retrieval step first.
Real entailment inference over (your context, the claim). A solved, scoreable problem: does the claim actually follow from what you supplied?
Real live Tavily search, then the same entailment model over the retrieved evidence. Consistent with the web right now — not a guarantee of objective truth.
No findable or relevant evidence → insufficient_evidence, always. Never rounded to a confident verdict to look decisive.
x-api-key for free, rate-limited access to POST /check and /check/batch — same free-tier model WickedAPI's main API uses.GET /checks/{id} retrieves a prior check's full result for dispute/audit, scoped to the caller who requested it.Optionally with context. POST /check for one claim, POST /check/batch for several against shared context.
supported, unsupported, contradicted, or an honest insufficient_evidence — plus confidence and cited evidence.
GET /checks/{id} pulls the full result back up for a dispute or a record.
Check a claim right now — free with an API key, or pay per call with x402:
curl -H "x-api-key: YOUR_KEY" -X POST https://sanity.wickedapi.com/check \
-d '{"claim":"...","context":"...","mode":"grounded"}'
Live, pulled straight from /v1/status — never out of sync with what's deployed.