Vercel adds TypeSafe AI's Jev 1.14 decision model to AI Gateway
Vercel has made Jev 1.14 from TypeSafe AI available on its AI Gateway. Jev is a decision model that returns typed Choice, Score and Boolean answers with probabilities instead of free-form text, and this version adds image input and an experimental long-context mode.
- Jev 1.14 from TypeSafe AI is now available on Vercel's AI Gateway under the name typesafe-ai/jev-1.14.0.
- Jev returns typed Choice, Score and Boolean answers with probabilities, and answers all questions in one parallel request.
- Version 1.14 accepts images as part of the state and adds an experimental long-context mode.
- Zero Data Retention and No Training can be enabled per request.
- Decision calls appear in logs and custom reporting and count toward budgets.
A decision model rather than a text generator
According to Vercel, Jev is a decision model for software. Developers pass it a state and a set of questions, and it returns typed Choice, Score and Boolean answers along with probabilities, not free-form text.
The model answers every question in a single request, in parallel. Vercel lists classification, routing, scoring and automated verification as the intended uses. The model name to use is typesafe-ai/jev-1.14.0.
Image input and long-context mode
Jev 1.14 accepts images as part of the state. Developers pass an image as a file part next to text. In the example Vercel provides, a local PNG is read from disk and sent with the text "A product photo from a support ticket." A Boolean question then asks whether the image is predominantly red.
The release also includes an experimental long-context mode. It is meant for documents, logs or conversation histories that are too long for a standard request. It is switched on with providerOptions.typesafe.longContext: true. Vercel's example reads a support log and asks a Boolean question about whether the log says the delivered item arrived damaged.
Both examples use the experimental_decide function from the AI SDK, imported as decide. The answers are read from result.answers.
Data controls and Gateway features
Vercel says Jev supports Zero Data Retention and No Training, and both can be enabled per request. Decision calls also appear in logs and custom reporting and count toward budgets. They accept other Gateway provider options in the same providerOptions.gateway object.
What it means for developers
For teams already routing model calls through AI Gateway, the release puts a model built for structured decisions next to their existing setup. Because answers come back as typed values with probabilities, application code can act on them directly, for example when routing a support ticket or verifying a condition, without parsing generated text.
Image input widens the kind of state a decision can be based on, and the long-context mode, which Vercel labels experimental, targets inputs such as long logs. Vercel points to its documentation on decision models on AI Gateway, a page of practical use cases for Jev, and a knowledge base for further details.
Vercel: TypeSafe AI's Jev 1.14 is now available on AI Gateway
