Jev Explained

# Jev Explained

**Learn how TypeSafe's Jev makes typed, probabilistic decisions — by running it.**

Live demo: **https://jev-explained-repo.vercel.app/** (bring your own TypeSafe or Vercel AI Gateway key).

<p align="center">
  <img src="docs/jev-primitives.png" alt="Jev primitives: Noul (yes/no), Choice (which one), Score (how much)" width="640">
</p>

An interactive playground that shows, step by step, how [Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) — TypeSafe's System One model — works.

## What is Jev?

Jev is not a chat LLM. It does not generate text. You send it a **state** (any text or JSON: an email, a market snapshot, a tool call an agent wants to run, a whole inbox) plus one or more typed **questions**, and it returns calibrated probabilities for every question in a single ~100 ms round trip. Your code, not the model, makes the final decision by thresholding on those numbers.

### The three primitives

Choose the primitive by the type of question you are asking:

| Primitive | Ask it when | Example | Returns |
| --- | --- | --- | --- |
| **Noul** — yes / no? | the question is binary | *Is this email spam?* | one probability, 0 → no, 1 → yes |
| **Choice** — which one? | you pick from known options | *Which team should handle this?* | the chosen option, a probability for every option, and a `confidence` |
| **Score** — how much / what level? | you grade on an ordered rubric | *How risky is this?* | a weighted score, a probability for every level, and a `confidence` |

Two things make this different from asking an LLM:

- **Questions run in parallel.** Jev reads the state once and answers every question at the same time, so ten questions cost about the same as one. You can fan out speculatively and let your code decide what matters.
-