What does “System One model” mean?

TypeSafe describes Jev as a System One model: a fast judgment layer for bounded decisions. The term contrasts with generative models that produce language token by token. A Jev request describes the state to evaluate and the questions to answer; the output is structured around those questions.

For example, a support system may send one customer message and ask which team should handle it, how severe the issue is, and whether a human should reply today. The application can then route, prioritize, or escalate using ordinary code.

The three Jev question types

Choice

Choice selects from a set of labels supplied by the application. Typical uses include support routing, intent classification, selecting an agent tool, or choosing a model tier. A probability distribution can help the application identify ambiguous cases.

Score

Score places the state on an ordered rubric, such as low, medium, high, and critical. The numeric result represents a position on that rubric; it should be interpreted using the exact response contract and labels in the current API documentation.

Noul

Noul returns a probability for a yes/no statement, such as “Does this need human review?” The application chooses its own thresholds and fallback behavior. A probability is an estimate, not a guarantee.

How Jev works in an application

  1. Represent the relevant text or program context as a state.
  2. Define one or more typed questions with clear criteria.
  3. Evaluate the returned values in application code.
  4. Set explicit thresholds and a fallback for uncertain or high-impact cases.
Typed output does not mean infallible output. A constrained answer can still be the wrong answer. Validate decisions against your own examples and keep human review for cases where an error matters.

What Jev is not designed to do

Jev is not a replacement for a conversational model when a task requires writing, summarizing, explaining, coding, or open-ended reasoning. A common architecture uses a decision model for routing or gating and a generative model for language tasks.

Sources and independence

This is an independent educational guide, not an official TypeSafe AI property. Product names and capabilities may change; consult the primary documentation before implementing an integration.

Next: Jev AI use cases →