Skip to main content

Jev Playground

Build TypeSafe System One requests — state plus Noul, Choice, and Score questions. Everything runs locally.

Your key stays in your browser — calls go direct to Jev

State — what Jev evaluates

Structured objects keep relationships explicit — questions can reference fields like incident.description.

Questions — typed, evaluated in parallel

noul
choice
score
L0
L1
L2

About the Jev Playground

Jev is TypeSafe AI's System One model: state and typed questions in, probability distributions out — no text generation, no parsing, no schema hallucination. Building requests by hand means getting three things right at once: the state object, the question instructions, and the per-primitive criteria shapes (a map of options for Choice, an ordered list of concrete levels for Score, optional yes-criteria for Noul).

This playground composes all three visually and shows a locally-generated mock response — full distributions with confidence — so you see the shape of what comes back before spending tokens. It then emits ready-to-run typesafe_sdk Python for your own terminal: the page never asks for an API key, because it never calls anything. Question IDs, field references like incident.description, and the vague-adjective warning on Score levels follow the documented guidance.

How to use

  1. Paste or build your state JSON — structured objects keep relationships explicit.
  2. Add questions: Noul (yes/no probability), Choice (fixed options), Score (concrete levels).
  3. Review the mock distributions to sanity-check your answer-space design.
  4. Copy the generated Python and run it from your terminal with your credentials.