Jev is TypeSafe AI’s first System One model. You send a state and one or more typed questions; it returns structured answers that application code can use. The three documented question types are Choice, Score, and Noul. This makes Jev relevant when software needs to classify, rank, evaluate, or route something, rather than produce prose for a person to read.

The interface in one example

Suppose a support ticket says a customer was charged twice. The ticket and any relevant policy form the state. One Choice question can select a team from a defined set; a Noul can ask whether the text requests a refund; a Score can rate urgency on a rubric. Your code then decides what to do with those results. The model supplies a judgment, while your application remains responsible for policy, arithmetic, permissions, and side effects.

Good first uses

  • Route messages into a small, named set of queues.
  • Score a passage against descriptive quality levels.
  • Check whether a claim is supported by supplied text.
  • Add a review path when a decision is uncertain.

These are examples of the officially documented primitives. They are not guarantees that a given application will perform well without evaluation.

Where it does not fit

Jev does not generate chat replies or code and is not a replacement for the language model inside a coding agent. The coding-agent guide recommends using a coding agent to build software that calls Jev for structured decisions. Jev also accepts text-based state, not image, audio, or video input, according to the state documentation.

For the documented jev-1.13 version, TypeSafe also warns about counting, numeric precision, date comparison, irrelevant context, and complex indirection. Keep deterministic calculations in code and test decisions on representative data. See the version-specific limitations before relying on a workflow in production.

Try the smallest useful question

Write down one decision your application already makes manually. Define its possible answers and the evidence needed to choose between them. Try it in the Playground, compare results on clear and ambiguous examples, then follow the API quick start.