What is Jev AI?

Jev is a System One Model developed by TypeSafe for fast structured decisions such as classification, routing and scoring.

What is Jev AI? The same model. People also search what is Jev, what is Jev in AI, and Jev model. TypeSafe is the company. Jev is the model. System One is the class of model. Keep those three distinct.

If you meant the company, see What is TypeSafe AI?. If you meant “is this ChatGPT?”, see Is Jev an LLM?.

What Jev is for

Use Jev when the hard part is a judgment and the next step is already code:

• Which queue should receive this ticket?
• Does this message ask for a refund?
• How severe is this bug report?
• Should this retrieved passage reach a generative model?

Jev is a poor fit when you need a paragraph, a patch, a summary, or a chain of tool calls. Those stay with an LLM.

The request in one picture

Every useful Jev call has two parts:

1. State — the evidence. A string, a JSON object, or an array of text.
2. Questions — one or more typed judgments about that state.

{
  "model": "jev-latest",
  "state": "Help! My payouts have been failing for 3 days.",
  "questions": {
    "is_urgent": {
      "type": "noul",
      "instructions": "Does this convey urgency?"
    }
  }
}

The official API returns answers under the same IDs, plus the versioned model that actually ran and token usage.

The three primitives

All three can share one request. Official docs say questions are evaluated in parallel against the same state.

Read the dedicated pages: Choice, Score, and Noul.

Who created Jev?

TypeSafe's launch post (15 September 2026) is signed Diogo Almeida, founder, TypeSafe. In that post he writes that he previously worked at OpenAI on instruction-following methods, and that Jev is named after William Stanley Jevons.

Treat third-party biographies and funding figures as unverified unless TypeSafe publishes them.

How Jev relates to TypeSafe and System One

TypeSafe
  → Jev
    → System One Model
      → structured decision
        → Choice / Score / Noul

System One is TypeSafe's name for models built to make fast, structured decisions that software can consume. The name refers to Kahneman's fast/intuitive System 1, not to a second product sitting beside Jev.

When last verified, the current public model ID was jev-1.13.0. Aliases jev-latest and jev-preview both pointed at that version. Pin a versioned ID after you tune thresholds.

What Jev is not

• Not a chatbot.
• Not official TypeSafe documentation. This site is independent.
• Not a fine-tuned-per-customer classifier. Official models docs say Jev is not fine-tuned or LoRA-adapted with customer data.
• Not a calculator. Official jaggedness notes tell you to keep arithmetic in code.

When to use Jev

Use it inside a backend or agent loop where you already know the legal answers. Define those answers in criteria, keep business rules in code, and treat confidence as a gate.

When not to use Jev

Do not use Jev to write the customer reply, invent categories it was not given, compare dates, or count items in a list. See limitations.

Common mistakes

• Treating a Noul of 0.5 as “medium intensity”. It means the model is split, not that the customer is moderately upset. Use a Score for intensity.
• Leaving out an other option on a Choice, then wondering why everything gets a department.
• Sending an entire CRM history as state. Official docs describe accuracy falling as unrelated state grows.

Next: How Jev works or make your first call. For HTTP details see the Jev API. For published rates see Jev pricing.