AI Gateway now supports TypeSafe clients and an HTTP API for Jev
Source
vercel.com
Author
Rohan Taneja
Date
Context
Vercel has expanded how applications can access Jev, TypeSafe AI's model for returning structured decisions and probabilities. Following its earlier Gateway launch, developers can use an existing TypeSafe client or call an HTTP endpoint, alongside the AI SDK integration. The update lets an existing client switch its base URL and access credentials while retaining its evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → calls and response format.
The HTTP route makes the same decision service accessible from other languages and frameworks. Requests through all three routes share Gateway billing and usage tracking, so teams can integrate Jev without moving those records outside their existing model gateway.
Key takeaways · AI-distilled
Existing TypeSafe clients migrate by changing only the base URL and API key - evaluation calls (state in, questions asked, response shape) stay exactly as they were before pointing at the gateway.
Jev returns typed answers with attached probabilities rather than generated text to parse, since it's built as a probabilistic decision model rather than a chat/completion model.
New integrations pick from three question types: boolean returns a 0-1 probability, choice picks one option from a named set, and score rates against a caller-defined scale.
eve, a framework for building agents with sandboxAn isolated environment where AI-generated code or agent actions run without being able to touch anything real.Full definition → compute and human approvals, uses Jev as its default evaluation model for automatic model selection and automated tool approvals.
Terms in this piece · Glossary
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
