Integration guideSources checked October 14, 2026
Jev on Cloudflare
You can use Jev on Cloudflare through Workers AI with the model ID typesafe/jev. Your Worker sends the information and typed questions, then uses the returned decisions in your application. This is TypeSafe’s Jev, separate from Cloudflare’s Clef models. Below, we explain the binding, request format, and response checks you need for a first integration.
What do you need to get started?
A Worker needs a Workers AI binding to call models through env.AI. Cloudflare lets you configure the binding in the dashboard or in the Wrangler configuration. Workers AI bindings
For a first test, keep the application small:
- Receive one text input.
- Prepare one clearly defined question.
- Call Jev through the AI binding.
- Display the result for inspection.
Do not begin by connecting the answer to an irreversible action. First make it easy to see what the model received and what it returned.
What should you send to Jev?
The request contains the information to evaluate in state and the decisions to make in questions.
For a simple prototype, imagine this input:
“I can open my account, but the download fails every time.”
You could ask which queue should receive it, with options such as account access, technical support, billing, and other.
This is a request-design example, not an observed Jev answer. The important part is defining the categories well enough that another person could apply them consistently.
From a Worker, Cloudflare’s documented call uses env.AI.run with the model ID typesafe/jev. Its Jev page includes examples for Choice, Score, and Noul, plus a REST example for calls from outside a Worker. Use the documented request format for the interface you choose. Jev request examples
How should your Worker use the response?
Inspect the answer associated with your question, then apply your application’s own rules.
For the prototype above, you might show the proposed queue next to the original message and let a reviewer correct it. Save those corrections as evaluation data. Once you understand the errors, decide which cases can be routed automatically.
Keep model uncertainty separate from request failure. A valid but uncertain answer is different from a timeout or an unavailable service. Your application needs a sensible response to both.
Is Jev on Cloudflare the same as Clef?
No. The typesafe/jev listing refers to TypeSafe’s model. Clef and Clef-flash are Cloudflare’s own decision models. Choosing one is a model-selection decision, even when the request formats are similar. Jev listing · Cloudflare’s Clef announcement
Use Jev on Cloudflare when you specifically want to evaluate Jev through that platform. Compare Clef separately if you are considering another model.
Common setup problems to check
| Symptom | First checks |
|---|---|
| Confirm the AI binding exists in the environment receiving the request |
| The request is rejected | Check the model ID and request shape against the current model documentation |
| Results are inconsistent with your labels | Review the input, category definitions, and ambiguous examples |
| Requests fail under load | Record the returned error and review the limits for your account and access route |
If your integration uses API credentials, keep them server-side. Cloudflare provides Worker secrets for sensitive values; do not place credentials in browser code. Cloudflare secrets documentation
What should you measure before launch?
For your Jev on Cloudflare test, record decision quality, total response time, failure rate, and actual usage charges.
Start with a narrow task and a visible review step. That gives you a working integration and evidence about whether the decisions are useful before you build a larger workflow around them.