Jev Playground
Build typed questions — Choice, Score, Noul — and run them against the live Jev API with your own API key. See exactly what your code would receive: answers, probabilities, confidence, tokens and cost. The key is stored in this browser only and sent directly to api.typesafe.ai — never to us.
1 · State
Plain text or JSON — all questions share this state and are answered independently in a single call.
2 · Questions
3 · Results
Or run it from your terminal
The exact same request as a shell command — useful if the browser blocks the direct call (CORS) or you prefer the terminal:
What just happened here
This page is a thin client for the official POST /v1/systemone endpoint. You supply the key; the request goes from your browser straight to api.typesafe.ai. We never see your key, your state, or the answers. Everything you type is autosaved in your browser's localStorage so a refresh does not lose your work.
One call carries one state and all of its questions — the same pattern you would use in production. The response mirrors it: one typed answer per question, each with probabilities and a confidence value your code can gate on.
The three question types
Choice — pick one of N options
2–255 named options, optionally with a description each. Best for routing and classification: "which team handles this?", "is this passage relevant or not?". The answer is the winning option plus a probability for every option.
Score — grade on an ordered scale
2–10 levels ordered low to high, starting at 0. Best for graded judgments: frustration, urgency, quality. Write the boundary of each level into its description — that is what makes scores comparable across inputs.
Noul — yes/no with a probability
The cheapest primitive: a single binary judgment, returned as the probability of yes. Optionally describe what yes and no mean. Use it for guardrails: "does this need a human?", "is this request sensitive?".
Limits, cost and honesty
- Billing is input-only at $0.042 per 1M tokens; output is free. This page shows the per-call token count and cost after every run.
- Limits: up to 64K tokens per request, 32K combined for state plus the longest question, ~1,200 requests/minute on direct access (subject to TypeSafe's current docs).
- A valid answer can still be wrong — Jev guarantees schema validity, not correctness. Gate actions on confidence bands: act automatically when high, review in the middle, escalate when low.
- Write atomic questions. One question = one judgment. "Score this startup" mixes market, tech and team — split it into several Score questions and combine in your code.
- No key yet? Registration is open with $5 in free credit — see Get access.