What is Jev?
Jev is the first System One model from TypeSafe AI. Instead of writing text like a chatbot, it reads your program's state and returns a typed, calibrated decision — a choice, a score, or a yes/no probability — that your software can act on directly.
The short version
Large language models were built to talk to people. Jev is built to be called by machines. You give it some context and the exact questions you need answered, and it replies with structured values instead of prose. Because the shape of the answer is fixed by your request, Jev cannot hallucinate a format or emit an invalid type — it can only return one of the options you defined, a score on the scale you gave it, or a probability between 0 and 1.
Why "System One"?
The name borrows from the psychology of thinking. "System Two" is slow, deliberate, verbal reasoning — that's what LLMs imitate when they write out an answer. "System One" is the fast, intuitive judgment that happens in a blink. Jev is designed for that role in software: the millions of small, quick decisions that shouldn't need a paragraph of generated text — routing, scoring, filtering, extracting, gating.
How Jev works: RLCD
Jev is trained with a method TypeSafe calls Reinforcement Learning for Calibrated Decisions (RLCD). The goal isn't fluent writing — it's calibration: when Jev says it's 80% sure, it should be right about 80% of the time. Every answer comes with a probability distribution, so your code can decide how much certainty a given branch requires and escalate the rest.
What you can ask it
Every question you send Jev is one of three primitives:
- choice — select one of up to 255 labelled options.
- score — place the input on an ordered 2–10 level scale.
- noul — a calibrated yes/no returned as a probability from 0 to 1.
You can ask several at once; Jev evaluates them in parallel in a single round trip. The best way to feel it is to run a live example in the playground.
Jev vs. a large language model
- Output: typed structured values & probabilities, not free text.
- Speed: 70–500ms — roughly 40–200× faster than a frontier LLM on comparable tasks.
- Cost: $0.042 per million input tokens, output free (≈ $0.0004 per decision).
- Safety: no hallucinated formats and no type errors, by construction.
- Use: called by other software and agents, not read by an end user.
Who built Jev?
Jev comes from TypeSafe AI, founded by former OpenAI researcher Diogo Almeida — a co-inventor of the RLHF work behind ChatGPT — together with Erik Gafni and Sasha Sheng. The company emerged from stealth in September 2026 with $40M in seed funding led by DCVC. Almeida's thesis: if AI is going to change how work gets done, people can't be the only consumers of intelligence — most of it should live quietly inside software.
Try Jev yourself
You don't need to join the waitlist to see what it does. Open the free playground, pick a real-world demo — support triage, moderation, lead scoring, code-review risk — and watch Jev return typed answers with calibrated probabilities. When you're ready to build, the API guide has copy-paste examples in curl, Python and TypeScript.