Running on jev-latest

See what Jev sees.

Jev is a System One model: it doesn't write prose, it returns typed decisions with calibrated probabilities. Paste any text below and watch nine of them come back in a single request — charted, shareable, and impossible to parse wrong.

~600 ms for nine answers $0.042 per million input tokens No free text to parse

Message X-ray

What is this message actually asking of you?

This becomes the state Jev reads. 0 / 12,000
Public

Appears in Explore and in search results.

Three question types

Everything Jev can answer is one of three shapes. Below, all three are answered against the same eleven-word state — the example straight out of the TypeSafe docs.

"Help! My payouts have been failing for 3 days."

Noul

Does this convey urgency?

Almost certainly
Probability the answer is yes
96%
No Coin flip Yes

A yes/no question comes back as a probability, not a word. 0.96 and 0.51 are different answers, and your code can tell them apart.

How a Noul works

Choice

Which team should handle this?

billing at 96%, from 3 options
96% billing

One option out of a set you define, with the full distribution behind it — so you can see the runner-up and how close it was.

How a Choice works

Score

How frustrated is the customer?

1.24 of 2 · Frustrated

A rating against ordered levels you write. The answer is probability-weighted, so it lands between levels when the text does.

How a Score works
1 request

Nine questions, one call. Jev reads the state once and answers every question against it in parallel.

0 parsing

Answers arrive typed. No JSON mode, no retries on malformed output, no regex over prose.

3 primitives

Noul, Choice, Score. Every decision your software needs is one of these three shapes.

± confidence

Choice and Score carry a confidence value, so "I am not sure" is a branch your code can take.

Coming from an LLM?

The same routing decision written both ways — and where the retry loop goes.

Fresh from the gallery

What other people have been pointing Jev at.

See all →

Typed decisions, all the way down

Dejev is a demo, not a product. The interesting part is the model underneath — read the TypeSafe docs to put Jev in your own software.