Vercel
Jev from TypeSafe AI is now available through Vercel AI Gateway, giving builders a model designed to make typed decisions rather than generate and parse text.
Jev takes application state as input and returns Choices, Scores and Booleans with probabilities. It evaluates all declared questions in parallel. Clear results can continue through an automated workflow, while uncertain results can be sent for manual review.
The Vercel changelog lists tool selection, risk scoring and guardrail enforcement among the possible uses.
How Jev from TypeSafe AI handles decisions
A regular language model generates text one token at a time. The application then needs to parse that text, check its structure and validate the answer before using it.
Jev removes that text-generation step. A workflow declares the questions it needs answered, then receives typed values and associated probabilities directly. Those outputs can feed into application logic without first converting prose into a usable value.
For example, an agent workflow could ask Jev which tool or subagent should handle the next task. A risk process could request a Score, while a guardrail could use a Boolean to decide whether an action is allowed. The probability attached to each answer provides a threshold for sending uncertain cases to a person.
TypeSafe reports the following results from its workflow evaluations:
- Speed
- Up to 193.6x faster
- compared with LLMs
- Cost
- Up to 444.6x cheaper
- compared with LLMs
These figures come from TypeSafe’s evaluations rather than an independent benchmark.
Using Jev through AI SDK 7
AI SDK 7 exposes Jev through its experimental evaluate API. Support starts with AI SDK version 7.0.105.
Builders can call the model through AI Gateway and define typed questions for the decision. The returned Choice, Score or Boolean can then drive the next workflow step, with probability thresholds used to separate automatic decisions from cases requiring review.
Jev joins Vercel’s wider set of AI Gateway models, but its job is narrower than a general-purpose language model. It is built for evaluating state and returning structured decisions, not drafting content or holding a conversation.
Frequently asked questions
What is Jev from TypeSafe AI on Vercel?
Jev is a probabilistic decision model available through Vercel AI Gateway. It evaluates application state and returns typed Choices, Scores or Booleans with probabilities instead of generating prose.
When is Jev from TypeSafe AI available?
Jev is available now through AI Gateway. AI SDK 7 supports it through the experimental evaluate API from version 7.0.105 onwards.
Which Vercel plan includes Jev from TypeSafe AI?
The changelog does not state which plans include Jev or give pricing details. Access is provided through Vercel AI Gateway.
Sources
3 checkedHow we cover tool news: Create With's tool desk drafts these reports with AI from the sources listed above and checks them against those sources before publishing.
n8n
