Jev & Laya: Typed Decision Models, API Guides and Open-Source Projects

Jev is TypeSafe AI's System One model: instead of writing prose, it answers typed questions — a choice, a score or a calibrated yes/no called noul — and returns probabilities with every answer. Laya is the open-source implementation of the same decision-model pattern, small enough to run on a laptop or an iPhone.

This site tracks both: how the API works, what the numbers actually mean, where the model fails, and which GitHub projects are worth your time. Everything here is written from primary sources — the TypeSafe API docs, the Laya model cards, and the public launch threads — and each page links back to them.

ReleasedSept 15, 2026
Made byTypeSafe AI
Answer typeschoice · score · noul
API modeljev-latest → jev-1.13.0
Open alternativeLaya, 421M params
Laya local speed~5 ms (Core ML)

Start here

Why a decision model instead of a chatbot

A chat model always produces a string, so every routing decision, severity score or moderation verdict has to be parsed back out of free text — and that parse is where pipelines break. A decision model removes the parse: you declare the questions once, and the answer comes back as a value with a probability attached. The trade is that you no longer get reasoning to read; if you need justification, you ask for it as another question, not as a sentence.

That single change is why the launch threads kept landing on the same line: AI executes, code decides. The model picks among options you defined; the arithmetic, the thresholds and the side effects stay in your code.

What this site is not

Not affiliated with TypeSafe AI, Perplexity, ConvAI Innovations or any project listed here. No API keys are sold or proxied. Numbers are quoted with their source and date, because both models ship fast — jev-latest already moved from one patch to the next inside a week, and Laya's checkpoints changed twice.

Last updated: 2026-09-21 · sources & corrections