Use Jev AI with Vercel AI Gateway

The Vercel route to Jev

Vercel’s September 16 announcement documents Jev through AI Gateway using typesafe-ai/jev. AI SDK 7.0.105 and later supports the experimental evaluation API. This is a structured decision call; the example does not ask Jev to write a chat response.

Use the provider route that fits your application. Gateway and direct TypeSafe requests have different model identifiers and some different response fields. Keep the relevant documentation beside your code.

A small server-side evaluation

Install a supported AI SDK version in a Node.js project:

npm install ai@^7.0.105

Set AI_GATEWAY_API_KEY in the server environment using your Gateway credentials. Then save this original example as evaluate.mjs:

import { experimental_evaluate as evaluate } from 'ai';

const result = await evaluate({
  model: 'typesafe-ai/jev',
  state: 'Please close my account after I export my saved reports.',
  questions: {
    requestsClosure: {
      type: 'boolean',
      instructions: 'Does the customer request closing the account?',
    },
  },
});

console.log(result.answers.requestsClosure);
console.log(result.usage);

Run node evaluate.mjs in that environment. This example logs a judgment; it performs no account action. It was checked against Vercel’s evaluation documentation but has not been executed by this site. No example probability or latency is claimed.

Gateway and direct TypeSafe: key differences

The TypeSafe API reference and Gateway evaluation reference define these contracts. For Choice and Score, the Vercel announcement describes TypeSafe confidence metadata under providerMetadata.typesafe.confidence.

If a copied snippet fails, compare its endpoint, model name, credentials, and question type as one set. Changing only the host while leaving a request from another interface can be the wrong adaptation. Vercel also documents a TypeSafe-compatible route; use its specific instructions if you need that compatibility.

The free Jev promotion has an end date

Vercel’s September 18 article advertised free Jev access on AI Gateway until September 25, 2026. Do not interpret that announcement as permanent free usage or assume it applies to a direct TypeSafe account.

For ongoing use, check the current model rate and account limits in Gateway. Our pricing guide explains the published TypeSafe rate separately.

Check the integration before expanding it

Our suggested first test set includes an explicit closure request, a question about closure, a message that quotes someone else’s request, and a request later withdrawn. Record the intended result before looking at model outputs.

Keep the judgment separate from the action: check permissions, dependencies such as an export, and your confirmation rules in application code. Save usage and failures along with successful calls so you can compare versions later. For a fuller evaluation plan, see benchmarks and limitations.

Sources checked September 23, 2026. This is an independent guide; access, packages, and provider terms can change.