Your AI is confidently wrong 15% of the time
Check catches hallucinations before they reach your users. Seven verification methods, one API call. You pick the speed/accuracy tradeoff.
Get your API key{
"content": "Patient should take
500mg aspirin daily",
"methods": {
"reasoning": 0.5,
"tool": 0.3,
"biprm": 0.2
}
}{
"verdict": false,
"confidence": 0.94,
"decision": "reject",
"reasoning": "Dosage exceeds
safe limits (max 325mg)"
}Seven methods, one API
Each catches different errors. You weight them based on risk tolerance.
Most teams start with reasoning + tool — under 3 seconds, catches ~95% of errors. Add biprm or ensemble when accuracy is non-negotiable.
How it works
POST your AI output with method weights
curl -X POST https://api.checkapi.ai/v1/verify \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{
"content": "Patient should take 500mg aspirin daily",
"methods": { "reasoning": 0.5, "tool": 0.3, "biprm": 0.2 }
}'Get back a verdict with confidence and reasoning
{
"verdict": false,
"confidence": 0.94,
"decision": "reject",
"reasoning": "Dosage exceeds safe limits (max 325mg)"
}Route on the decision
if (result.decision === "reject") blockResponse();
if (result.decision === "escalate") routeToHuman();
if (result.decision === "accept") serve(result);Full docs at /docs — Python, Node.js, and cURL.
Built by
Drove $35M revenue solutions at Mastercard. Built $19M RFP bundles covering risk and compliance at West Monroe.
AI Software Engineer at Epoch, building autonomous coding agents. Previously built $5M DOD-funded sensor systems with zero downtime.
We spent years building AI systems at production scale. The hardest part wasn't making AI smart—it was trusting its outputs.
Pricing
Bring your own API keys. You pay for verifications, not our margins.
- 25,000 verifications/month
- All 7 methods
- 200 requests/min
- Batch processing
- Up to 5 team members
- 30-day data retention
- Unlimited verifications
- All 7 methods
- 1,000 requests/min
- Batch processing
- Unlimited team members
- 90-day data retention