How to score OpenAI answers with ClearRun

Get the answer from OpenAI as usual, then pass the prompt and the answer text to clearrun.evaluate(). ClearRun scores whether the answer's confidence is backed by its reasoning.

Last updated

Steps

  1. Install both

    npm install openai @clearrun/sdk

  2. Get an answer, then score it

    import OpenAI from 'openai';
import { createKernelClient } from '@clearrun/sdk';
const clearrun = createKernelClient(); // reads CLEARRUN_API_KEY

const openai = new OpenAI();
const prompt = 'When was the Eiffel Tower finished?';

const completion = await openai.chat.completions.create({
  model: process.env.OPENAI_MODEL,
  messages: [{ role: 'user', content: prompt }],
});
const answer = completion.choices[0].message.content ?? '';

const { verdict } = await clearrun.evaluate({ prompt, responseText: answer });
console.log(verdict.score, verdict.risk, verdict.verdictId);

Good to know

  • The answer must be at least 10 characters for evaluate.
  • Each evaluate call uses ClearRun credits, separate from your OpenAI bill.
  • Run this on a server. Both keys must stay out of browser code.