Deterministic fallback
Deep audits run via the shared Kimi proxy (kimi-k2.7-code). Any AI failure
falls back to keyword heuristics — the service never raises.
Agents say “all tests pass” without running them. VerifyAI reads the run transcript, checks each claimed action against what actually executed, and seals the verdict to a hash-chained vault.
Coding agents report “all tests pass”, “file edited”, “deployed to production” — without ever running the command. Teams merge on trust, and the audit trail is a chat log nobody can verify.
One pass over the run transcript. Every claim gets a verdict, a written reason, and a confidence score — then the whole audit is hashed into the vault.
Paste the agent transcript and its claimed actions —
tests passed, file edited, deployed.
ClaimCheckAI checks every claim against execution markers —
pass counts, diff --git, build success, 200 OK.
Each claim returns verified,
partially_verified or unverified with a reason
and confidence.
The sha256 of the audit is chained into the evidence vault and the run lands on your verdict wall.
$ pytest tests/ -q
..............................
42 passed in 3.14s
# note: no deploy step was ever run
Press Run the audit to see per-claim verdicts.
Every claim is verified, partially_verified or
unverified, with a written reason and a confidence value. No stars, no percentages
dressed up as judgment.
Deep audits run via the shared Kimi proxy (kimi-k2.7-code). Any AI failure
falls back to keyword heuristics — the service never raises.
Each audit is serialized and hashed into the platform vault as
kind='verifyai_audit'.
Heuristic engine, per-claim verdicts, verdict wall and sealed hashes — no card required.
POST /v1/runs submits a transcript and claim list;
GET /v1/runs/{id} returns full verdict detail. Drop it in a pipeline step and
fail the build on unverified deploy claims.
A verdict you can prove months later. Every audit seals its result to the platform evidence vault. Records are hash-chained: change one byte of a past verdict and every seal after it breaks.
Start free. Upgrade when your agents start lying more often than your schedule allows.
Running a platform? The API tier at $199/mo adds dedicated throughput and CI webhooks — talk to us from the app.
Submit your first agent run in under a minute. Fifty checks a month, free — every verdict sealed.