Jev

System One decision model

Developer guide

What is Jev AI? TypeSafe System One Model Guide

Jev is the System One decision model introduced by TypeSafe AI. It is designed for AI agent pipelines where predictable, consistent decisions matter more than creative text generation. The core design philosophy is simple: Decisions not strings.

General-purpose large language models are optimized for writing, reasoning and natural conversation. They are powerful, but their outputs can be unpredictable when used inside automated agent workflows. Jev solves this gap by delivering structured, repeatable decision results.

Access & Use Jev Watch Jev videos See Jev feature cards See speed & cost charts

Watch: how Jev works and how to use it

Short explainers for What is Jev AI?, how to use it, and what you can build. Play on this page (needs YouTube access).

JEV: How It Works and What You Can Build

Riley Brown

Jev is HERE. How to use it

Greg Isenberg

Jev Explained: Demos and Use Cases

Syntax

Jev - The Ultimate Classification Model?

Sam Witteveen

Prefer reading? Start with the System One explainer or Access & Use Jev.

Why teams pick Jev: speed & cost

Charts based on published TypeSafe ranges. Full report with methodology and scenario templates lives on the benchmarks page.

Chart comparing Jev latency of 70 to 500 milliseconds against typical frontier LLM ranges of about 3 to 329 seconds — Latency — Jev’s published System One band stays in milliseconds while frontier chat models often sit in seconds.

Chart comparing Jev input price of 0.042 dollars per million tokens against higher LLM price bands — Input price — about $0.042 / MTok, with published free output tokens for decision results.

Chart showing about 194 times faster for Jev on workflow evaluations — Workflow speed headline — ~194× faster on published workflow-style evals.

Chart showing about 445 times cheaper for Jev on workflow evaluations — Workflow cost headline — ~445× cheaper on the same published eval framing.

Four-panel scenario matrix for routing, batch scoring, real-time loops, and high-cardinality choice — Test scenarios — templates for routing desks, batch scoring, real-time loops, and large Choice sets.

Open the full benchmarks report

Jev features at a glance

Six visual cards covering the talking points that show up again and again in public Jev intros: decisions instead of chat text, smart if-style routing, Choice / Score / Noul, parallel questions, calibrated confidence, and schema-safe outputs.

Dig into the primitives in the SDK guide or the glossary.

System One vs System Two

The System One / System Two naming comes from behavioral psychology. In agent engineering terms, the split maps cleanly to two different jobs inside one pipeline. See the glossary for short definitions.

System One (Jev): Fast, deterministic, consistent classification and decision making. Low variation, ideal for automated agent workflows.
System Two (Generative LLM): Slow, flexible reasoning, text creation, open-ended conversation. Good for writing and complex reasoning.

Production stacks rarely pick only one. Most teams put Jev at the decision edge, then call a generative model only when the chosen path needs prose, tool planning, or deeper reasoning. Read the comparison page for a feature-by-feature breakdown.

Quick navigation

Official sources & further reading

Product claims and brand assets belong to TypeSafe AI. Start with the official announcement, or read an independent launch summary on Truescho.

Source references: