What is Jev AI?

Jev AI is a decision model for structured judgments.

Jev is an AI model from TypeSafe AI for bounded judgments inside software. You supply the state, the question, and the allowed answer shape. It returns typed decisions and probabilities—not prose.

TypeSafe calls this a System One model. In practical terms, it can route a support ticket or score its urgency. It cannot draft the reply, investigate the issue, or decide what actions your software permits.

See one request Use the API

State + typed questionsJevChoices, scores, probabilitiesYour application

The useful boundary

Use it for a branch, not a blank page.

Jev is narrow by design. That makes it useful when your software needs a judgment with a known destination.

Jev fits when code needs to…

route a request to one of several queues
score severity against a defined rubric
estimate whether one proposition is true
judge an agent result before the next step

Use something else when you need to…

write, summarize, explain, or generate code
plan or investigate across several steps
apply a rule that ordinary code can express exactly
decide from facts missing from the supplied state

See the Jev vs. LLM architecture comparison

The output contract

Three question types, each with a different answer.

Choice

Pick one item from a closed, unordered set.

choice + probabilities + confidence

Score

Place the state on an ordered rubric you define.

score + distribution + confidence

Noul

Estimate the probability that one statement is true.

P(True), from 0 to 1

Read the complete primitive reference

One important caveat

Typed does not mean correct.

A Choice answer cannot invent a label outside your schema. It can still choose the wrong valid label. Jev’s output contract removes free-form responses; it does not remove model error.

Test decisions on your own data. Keep permissions, thresholds, fallbacks, and irreversible actions in code.

Separate guarantees from claims

The wire format

A Jev call is state plus named questions.

This request asks two independent questions about one support ticket. The response values below are illustrative; the schema is the point.

Request / cURL

curl -s https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "I was charged twice this month. Please refund the duplicate.",
    "questions": {
      "team": {
        "type": "choice",
        "instructions": "Route this support ticket",
        "criteria": {
          "billing": "Charges, invoices, or refunds",
          "technical": "Errors or product failures",
          "other": "Neither category fits"
        }
      },
      "refund_requested": {
        "type": "noul",
        "instructions": "The customer explicitly asks for a refund."
      }
    }
  }'

Response / illustrative values

{
  "answers": {
    "team": {
      "type": "choice",
      "choice": "billing",
      "confidence": 0.92,
      "probabilities": {
        "billing": 0.92,
        "technical": 0.02,
        "other": 0.06
      }
    },
    "refund_requested": {
      "type": "noul",
      "noul": 0.97
    }
  }
}

Get runnable SDK examples and access options

Go deeper