jev

Jev is TypeSafe AI's "System One" decision model, in early access since 2026-09-15, reachable on the Vercel AI Gateway as typesafe-ai/jev. It does not generate text: you send a state (text or JSON) plus a set of typed questions, and it returns one answer per question — a probability, a picked option, or a score — all evaluated in parallel in a single call. It is trained with RLCD (Reinforcement Learning for Calibrated Decisions), so a reported confidence is meant to mean what it says: an answer given 90% probability should be right about 90% of the time.

Use Jev when / do NOT use Jev when

All three must hold, or it is the wrong tool: