CalibratedAgents
Try in console

API

Check every claim, from your code.

One HTTPS call checks an AI answer against the passages your system retrieved. You get one calibrated score for the answer, and for every wrong claim, why, where in your sources and the fix.

At a glance

Base URLhttps://api.calibratedagents.com
AuthenticationAuthorization: Bearer $CA_API_KEY — create a key in the console → API keys
FormatJSON in, JSON out, over HTTPS
Free allowance1,000,000 tokens a day · no card
Specification/api/openapi.json · OpenAPI 3.1
For AI coding agents/llms.txt lists every page

Endpoints

EndpointWhat it does
POST /v1/ground/basicCheck one answer: one calibrated score, and for every wrong claim, why, where and the fix.Docs →
POST /v1/ground/basic/batchCheck several answers in one call.Docs →
GET /v1/healthIs the API up? Returns ok and the time it was checked.Docs →

Your first call

Create a key in the console, set it as CA_API_KEY, then:

curl
curl -s https://api.calibratedagents.com/v1/ground/basic \
  -H "Authorization: Bearer $CA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "Can I prepay my personal loan after six months?",
       "context": ["Prepayment is allowed after 12 EMIs."],
       "answer": "You can prepay after 6 EMIs."}'

The claim scores about 0.97 — confidently wrong — and its receipt says 12. What the response means →

Everything for developers

Questions, answered.

What is the API's base URL?

https://api.calibratedagents.com. Authenticate with Authorization: Bearer and an API key from the console; requests and responses are JSON.

Is there an OpenAPI specification?

Yes: /api/openapi.json (OpenAPI 3.1). AI coding agents can also start from /llms.txt, which lists every page.

Is there a free tier?

Yes: 1,000,000 tokens a day, with no card.

How should my code handle errors?

Treat any non-200 response as no verdict. The API fails closed: if a check can't run you get a 503, never a silent pass.

Give us a week of your logs. We'll show you what your AI got wrong.