Which one?
Pick from options you define.
billing | technical | sales | other SMALL JUDGMENTS. REAL APPLICATIONS.
Not another chatbot.
An AI that decides.
Jev understands language and returns answers your code can use. A choice. A yes/no probability. A score.
No generated paragraphs. Just the judgment you need.
LANGUAGE IN
“I was charged twice.
Can I get my money back?”
↓ Jev understands the intent
{
team: "billing",
refund_requested: 0.98
}
team: "billing",
refund_requested: 0.98
}
STRUCTURED ANSWERS OUT Illustrative
01 / THE IDEA
Give your code a little common sense.
Jev is TypeSafe’s first System One model: built for focused judgments, not writing or long-form reasoning.
01
→Give it context
A message, a document, or structured text. This is the state.
02
→Ask a narrow question
Define what to judge and what the possible answers mean.
03
Let code take it from there
Route, rank, flag, or ask a human. Your application stays in control.
02 / THREE BUILDING BLOCKS
Is it true?
Return the probability of yes.
refund_requested → 0.98How much?
Rate against descriptive, ordered levels.
severity → 1.94 on a 0–2 scale03 / LEARN BY TRYING
Small decisions.
Lots of possibilities.
Pick an application. Follow the path from language to a useful action.
2. THE JUDGMENT
Video-editing performancePortability
3. YOUR CODE DECIDES
Demo policies use 0.70 confidence for Choice/Score, 0.20/0.80 Noul cutoffs, and 1.50 for high scores. These are teaching defaults, not validated production thresholds.
See the API request & response
REQUEST
RESPONSE
THE TAKEAWAY
04 / THE IMPORTANT PART
Jev judges.
Your code runs the show.
Not a text generator. Pair it with a generative model when you need a written answer.
Not a calculator. Keep arithmetic, dates, permissions, and side effects in code.
Not infallible. Typed answers guarantee shape, not truth. Validate on your data and send uncertain cases for review.
One state, many questions. Batch independent judgments together. They run in parallel, without seeing each other’s answers.