jevx-client

A small typed client for the Jev AI decisions API on Jevx. Send a state and the questions you branch on, get back the choice, a probability on every option and a confidence. Zero runtime dependencies, Node 18+.

Jev AI is a zero-shot decision model from TypeSafe: text or JSON in, typed values out. This package is an independent client for the Jevx API and is not affiliated with TypeSafe AI.

npm install jevx-client

Create an API key at jevx.org/settings/apikeys and export it as JEVX_API_KEY.

import { decide } from 'jevx-client';

const { answers, charged } = await decide({
  state: 'Customer wrote: my order arrived three days late and the box was crushed.',
  questions: {
    sentiment: {
      type: 'choice',
      instructions: 'Overall sentiment of the customer.',
      criteria: { positive: 'Happy.', neutral: 'Neither happy nor upset.', negative: 'Unhappy or angry.' },
    },
    needs_human: {
      type: 'noul',
      instructions: 'Does this need a human agent?',
      criteria: { true: 'Requires a human agent.', false: 'Can be handled automatically.' },
    },
  },
});

if (answers.sentiment.type === 'choice' && answers.sentiment.probabilities.negative > 0.8) {
  await escalate();
}

Question shapes

| type | criteria | returns | |---|---|---| | choice | object of option to meaning, two or more options | choice, probabilities, confidence | | score | ordered array, low to high | score, probabilities, confidence | | noul | exactly the keys true and false | noul between 0 and 1, where exactly 0.5 means the model declined |

validateQuestions() runs the same shape checks locally and decide() calls it before sending, so a malformed question fails before it reaches the API. The API reports failures as HTTP 200 with a non-zero code; decide() turns those into a JevxError.

Up to 16 questions per request. The playground and the full guide are at jevx.org.

License

MIT