Dejev

How it works

Dejev is a thin, visual wrapper around one API call. Everything on a result page comes from a single POST /v1/systemone.

The whole request

You supply the state — whatever you paste. The lens supplies the questions. Jev returns one typed answer per question, keyed by the same ids.

{
  "state": "Help! My payouts have been failing for 3 days.",
  "model": "jev-latest",
  "questions": {
    "is_urgent":  { "type": "noul",   "instructions": "Does this convey urgency?" },
    "department": { "type": "choice", "instructions": "Which team should handle this?",
                    "criteria": { "billing": "Payments, invoicing, refunds",
                                  "technical": "Bugs, outages, integrations",
                                  "sales": "Pricing, upgrades, new accounts" } },
    "frustration":{ "type": "score",  "instructions": "How frustrated is the customer?",
                    "criteria": ["Calm", "Frustrated", "Very angry"] }
  }
}

How the charts map to the answers

Choice → donut
The probabilities sum to one, so a ring is honest here: it is a genuine part-to-whole. The legend repeats every option with its exact value, because a donut is bad at close calls and the numbers are not.
Score → rubric strip
The strip is your levels in order; the needle is the probability-weighted value, which is why it usually lands between two of them. Bars below show how the probability mass is spread — a score of 2.0 from a tight peak and a score of 2.0 from two distant humps are very different answers.
Noul → meter
One ratio against a limit, so it gets a meter rather than a chart. The grey band around the middle marks where the model is effectively saying "I don't know" — a 0.51 should never be read as a yes.

The lenses

Each lens is just a fixed set of questions chosen to show a different part of what Jev can do. You can also write your own.

  • Message X-ray

    What is this message actually asking of you?

    9 questions

  • Support triage

    Route a ticket before a human ever opens it.

    9 questions

  • Guardrail

    Screen a prompt before it reaches your model.

    9 questions

  • Draft critique

    Nine reads on a draft, before you send it.

    9 questions

  • Review decoder

    What a customer review is really telling you.

    9 questions

  • Bug report triage

    Severity, component, and whether anyone can reproduce it.

    9 questions

  • Headline lens

    How a piece of news is being framed.

    9 questions

  • Pitch check

    Would an investor or a customer keep reading?

    9 questions

Questions

What is a System One model?
A model built for decisions rather than prose. Jev takes one state plus a map of typed questions and returns one typed answer per question — no tokens streamed, nothing to parse, and no way for it to emit a value outside the set you defined. TypeSafe's framing is System One for fast structured judgement, with an ordinary LLM as System Two when you actually need language.
Why does everything run in a single request?
Jev ingests the state once and evaluates every question against it in parallel. Ten questions in one call cost roughly one call’s worth of input tokens instead of ten, and take about as long as one. That is why every lens here asks nine things at once instead of chaining prompts.
What does confidence mean?
Choice and Score answers carry a confidence value derived from their probability distribution: concentrated means certain, spread out means the model cannot separate the options. It is a second axis — the answer tells you what, confidence tells you whether to act. Noul answers have no confidence field, because the probability itself already carries it: 0.51 is the model saying it does not know.
Can Jev hallucinate?
Not in the shape of its output. A Choice can only ever return one of the options you defined; a Score can only land inside your rubric; a Noul is a number between 0 and 1. It can still be wrong about your text — which is exactly what confidence is for.
Why is it called Jev?
After William Stanley Jevons, the economist. TypeSafe’s launch post says they expect machine intelligence to follow the path coal took after steam engines got efficient — the Jevons paradox, where making something cheaper to use increases how much of it gets used rather than less.
Can I run Jev locally? Is it open source?
No. Jev is a proprietary hosted model reached over TypeSafe’s HTTP API; there are no published weights to download, and the same weights serve every account — it is not fine-tuned or LoRA-adapted per customer. You shape its answers through the request instead: what you put in the state, and how you write each question’s instructions and criteria.
How fast is Jev?
TypeSafe quotes 70–500 ms end to end, against 3–329 seconds for frontier models, and 40–200× faster on the kind of query it is built for. Those are the vendor’s figures. What this site can say from its own logs is that a nine-question request typically comes back in a few hundred milliseconds — every result page shows the measured round trip for that particular call.
What does Jev cost?
$0.042 per million input tokens on jev-1.13.0, with output free — there is no autoregressive decoding to meter. Because the state is read once and every question is evaluated against it in parallel, asking ten things costs roughly what asking one costs. Each result page on this site prints the token count and the cost of that exact call.
How was Jev trained?
With what TypeSafe calls RLCD — Reinforcement Learning for Calibrated Decisions. The stated goal is different from RLHF or RLVR: not answers people prefer or answers that pass a checker, but answers whose probabilities are epistemically honest. That is what makes confidence worth thresholding on rather than decorative.
What are the limits?
Text in only — no images, audio or video. 64k tokens per request, of which the state plus the single longest question must fit in 32k. A Choice takes up to 255 options; a Score takes 2 to 10 levels. Rate limits on early access are 250,000 tokens per second and 1,200 requests per minute, and TypeSafe warns they move.
What happens to what I paste?
It is sent to TypeSafe for evaluation and stored in this site’s database so the result has a permanent link. Public runs appear in Explore and in search engines. Unlisted runs are stored too, but are kept out of Explore and the sitemap and are served with a noindex header — the link still works for anyone you give it to. Either way, do not paste anything sensitive: this is a public demo.
Is this run by TypeSafe?
No. Dejev is an independent demo built to show what Jev’s output looks like when you draw it. The model, the API and the docs are TypeSafe’s.

Go deeper

The model, the pricing, the jagged edges and a long shelf of cookbooks are all in TypeSafe's own documentation.

Jev vs an LLM TypeSafe docs API reference