Glossary: System One
System One is TypeSafe’s name for a class of models that return software-consumable decisions (Choice, Score, Noul) instead of free-text completions. The name nods at Kahneman’s System 1: fast, focused judgments a knowledgeable person could make in a few seconds. Jev is the first public instance. The launch post (15 September 2026) frames Jev as unstructured state in, typed probabilistic decisions out.
This glossary entry is for builders who already heard of Jev / System One and need a citable definition before they treat it as a chat wrapper they can download. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Confirm the concept page on docs.typesafe.ai. We do not sell or issue API keys.
How it shows up
You do not send system_one as a model id. You POST https://api.typesafe.ai/v1/systemone with model set to an alias or versioned id such as jev-latest or jev-1.13.0. SDK methods are named system_one / systemOne. State is text/JSON only. Question IDs are not used in inference. Choice: unordered, max 255, choice + probabilities + confidence. Score: 2–10 levels, fractional score allowed. Noul: P(yes) 0–1, no confidence.
TypeSafe describes parallel evaluation of questions rather than token-by-token generation. Extra questions cost tokens and, per docs, barely change latency — a vendor description, not an SLA from this site. Training story: RLCD (named, not fully specified). Calibration is group-level, not a per-row guarantee. Schema-safe ≠ factually correct. Official docs say System One models do not write replies, produce code, or generate explanations.
TypeSafe documents a System One LLM wrapper they use for comparisons. That wrapper is not Jev, and they say it is slower and more expensive than the dedicated model — vendor claim.
Contrast
System One vs LLM chat. Chat emits text. System One fills typed slots. Code owns control flow.
System One vs Jev. System One is the class and the HTTP family. Jev is the public model you call today.
System One vs Kahneman System 2. The name is a nod, not a claim that Jev does slow deliberative proof. TypeSafe writes that it believes System One models can be made more reliable than the alternatives and defers the longer argument.
System One vs jev-latest. The alias picks which Jev checkpoint. The path stays /v1/systemone.
Common mistakes
- Sending
system_oneasmodel. - Expecting generated emails, code, or rationales.
- Treating the comparison LLM wrapper as Jev.
- Assuming jev.pro is official System One docs.
- Using a Noul of 0.5 as a “medium System One score.”
See also
FAQ
Is System One an LLM wrapper we can download?
TypeSafe documents a System One LLM wrapper they use for comparisons. That wrapper is not Jev, and they say it is slower and more expensive than the dedicated model.
Is jev.pro official System One docs?
No. Canonical docs live at docs.typesafe.ai.
Do I pass system_one as the model?
No. Pass jev-latest or a versioned id such as jev-1.13.0. The path is /v1/systemone.
Does System One guarantee each answer is correct?
No. Calibration is described across groups of predictions.
Limits
This is a jev.pro working definition. System One is TypeSafe’s product name for the class and API. We do not invent search volume. Schema-safe answers are not automatically factually correct. Latency and wrapper-cost comparisons are vendor claims.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.