How to access the Jev API

As of September 22, 2026, three public routes are worth evaluating: TypeSafe's native API, OpenRouter, and Vercel AI Gateway. They expose the same core idea — state plus typed questions — but differ in onboarding, billing, model names, and client compatibility.

1. TypeSafe direct

Use the native route when you want the model creator's own API and documentation.

• Start with the official quickstart.
• Request or manage access in the TypeSafe console.
• Use an official JavaScript SDK or Python SDK.

The official launch described Jev as early access. Check the console for the current access state rather than assuming approval is immediate.

2. OpenRouter

OpenRouter lists TypeSafe's Jev family and publishes current provider information on its TypeSafe page. It is useful if your team already centralizes model access and billing through OpenRouter.

Jev is a decisions model, not a chat-completions model. Do not assume a normal OpenAI-compatible chat endpoint is the correct path. Follow OpenRouter's current Jev documentation and model page, including its dedicated decision endpoint and model identifier.

The rolling Jev Latest alias follows the latest model in the family. A rolling alias is convenient for exploration; a pinned version is safer when reproducibility matters.

3. Vercel AI Gateway

Vercel supports Jev through three paths:

• its AI SDK evaluation API;
• a native HTTP evaluation endpoint;
• an existing TypeSafe client pointed at the gateway.

The Vercel Jev model page shows the current model entry. Vercel's TypeSafe client and HTTP API guide includes working examples and current endpoint details.

This route is attractive when an application already uses Vercel's billing, logs, budgets, or AI SDK. It is not automatically better than the direct API; it is an operational choice.

Which route should you choose?

Choose TypeSafe direct when you want the canonical API and official SDK behavior.

Choose OpenRouter when your model access and billing are already consolidated there, or when you want to compare Jev versions through its catalog.

Choose Vercel AI Gateway when your TypeScript application already uses AI SDK or you want Jev calls to share Vercel observability and budgets.

Production checklist

Before shipping any route:

1. Pin the provider, endpoint, and model identifier you tested.
2. Verify the accepted question types and response fields against current docs.
3. Set a deadline and treat overload or authentication failures as explicit application states.
4. Keep API keys on the server.
5. Test thresholds against labeled examples from your own workflow.
6. Log request IDs and safe configuration details, never secrets or private raw content.

A successful HTTP response proves only that the provider accepted the request. Production acceptance means the typed answer was parsed, the intended policy was applied, and an uncertain result followed the correct review path.