An honesty score for
every AI answer.

Pricing

ClearRun checks whether the confidence in an AI answer matches the justification behind it. Scores use credits.

• Open SDK: free. Run it in your own stack.
• Free account: 30 credits a month, set up the first time you sign in.
• Pro: from $25/month for 100 credits, up to $2,250/month for 10,000 credits.
• Business and Enterprise: talk to us.

Use it where you already work.

ClearRun scores honesty (confidence-justification alignment). It is not a fact-checker. Get an API key to call the hosted API.

{
  "mcpServers": {
    "clearrun": {
      "command": "npx",
      "args": ["-y", "@clearrun/mcp"],
      "env": {
        "CLEARRUN_API_KEY": "YOUR_CLEARRUN_API_KEY"
      }
    }
  }
}

npm i -g @clearrun/cli
export CLEARRUN_API_KEY=YOUR_CLEARRUN_API_KEY
clearrun score --text "The capital of France is Paris."
clearrun score answer.txt

import { createKernelClient } from '@clearrun/sdk';

const client = createKernelClient({
  hosted: { apiKey: process.env.CLEARRUN_API_KEY },
});

const scored = await client.score({
  responseText: 'The Eiffel Tower is in Berlin.',
});

clearrun verify VERDICT_ID
curl https://api.clearrun.net/api/verify/VERDICT_ID

Version 0.2.0 adds clearrun login, so you can sign in once instead of setting a key.

Use it as a checkpoint between AI steps.

Put ClearRun between the steps of your own pipeline or agent. Each answer gets an honesty score before it moves to the next stage, and your code sets the rule: pass it on, send it back for another try, or route it to a person when the score is below the level you choose. ClearRun never blocks anything itself. You set the threshold.

import { createKernelClient } from '@clearrun/sdk';

const client = createKernelClient({
  hosted: { apiKey: process.env.CLEARRUN_API_KEY },
});
const { verdict } = await client.score({ responseText: priorAnswer });
if (verdict.score == null || verdict.score < 70) {
  // retry or flag for a person
} else {
  // continue to the next stage
}

Three things, every answer.

See it work.

Two minutes: a question goes in, the score comes back, the flag opens, the audit prints.

Get early access.