Glossary

Jev & System One Model Glossary

Need a walkthrough instead of definitions? Start with the getting started guide or the System One explainer.

System One Model

Specialized AI model category built for fast deterministic decisions, not text generation. Jev is TypeSafe AI’s System One model.

Decisions not strings

Jev core design principle; model outputs structured decisions instead of unstructured natural language text.

Choice

Jev primitive that selects one option from a predefined list you supply. Common for agent routing and classification.

Score

Jev primitive that places state on an ordered rubric or graded scale. Common for risk, severity, and quality scoring.

Noul

Jev primitive that returns a yes/no probability judgement. Common for moderation gates and boolean filters.

Agent routing

The task of directing user requests to the correct sub-agent or tool inside multi-agent systems.

Deterministic output

Given identical input, the model returns consistent results.

System Two

Common shorthand referring to generative LLMs optimized for natural language generation and open reasoning.

Hybrid agent stack

Architecture combining System One decision models (Jev) + generative System Two LLMs inside one agent pipeline.

API-only (no local weights)

Jev is delivered as a hosted API. Public downloadable model weights for local self-hosting are not available.