What is Jev?

01It doesn't write text

Jev is TypeSafe AI's System One model — named after Kahneman's fast, intuitive "System 1" thinking. It's not a chatbot and it doesn't generate prose. You send it the state of your program plus a typed question, and it sends back one typed answer with a calibrated probability.

Three question shapes cover everything:

• Choice — pick one option from a set you define
• Score — place the state on an ordered scale (2–10 levels)
• Noul — a yes/no question, answered as a probability

02Why it's fast and cheap

Jev answers a full batch of typed questions in one parallel pass, typically in 70–500ms. TypeSafe reports 20–200x faster and 40–400x cheaper than a conversational LLM on the same classification task — input is priced at $0.042 per million tokens, output is free (there isn't any).

That comes from a purpose-built stack: a model architecture and parallel sampler built for structured decisions rather than open-ended generation, trained with a method TypeSafe calls RLCD — Reinforcement Learning for Calibrated Decisions.

03Where it fits

Jev sits next to an LLM, not in place of one. Common uses: routing a request to the right model or handler, moderating content in real time, scoring leads or tickets, reranking search results, verifying a tool call before it runs, and the inner loop of an agent or game AI that has to decide dozens of times per task.

The pattern TypeSafe calls out: code does the math (spelling, arithmetic, lookups), Jev makes the call.