Meet Jev, TypeSafe’s System One model
A new interface for decisions inside software.
A model with a different job
Jev is built by TypeSafe AI for structured judgments inside software. The team announced early access on September 15, 2026. Rather than asking for a paragraph and parsing it, you define the decision you need to make.
Its three decision primitives are Choice, Score and Noul. Since the October 14 Jev 1.14 release, experimental image inputs and a fourth question type, Box, are also available. Read the multimodal update for the current capabilities and opt-in requirements.
Start with a narrow decision
Consider a support inbox. “Handle this customer” is an open-ended task. “Which team should receive this request?” has a smaller, explicit set of outcomes. That boundary makes the result easier to evaluate and use.
Your code remains responsible for the next action. A model can help select a route; it should not silently redefine your permissions, refund rules or escalation policy.
Keep uncertainty visible
Type safety is about the output contract. It is not a promise that every judgment is right. Evaluate representative cases and decide how uncertain results should reach a person.
This is an independent introduction from jevai.me. Read the TypeSafe announcement for the provider’s claims, methods and limitations. To see the shape of each primitive, open our decision explorer.