Concepts
What is a System One Model?
What is a System One model? Jev is TypeSafe's first System One model: fast structured decisions for software, trained with RLCD.
- Published
- Sep 20, 2026
- Updated
- Sep 20, 2026
- Last verified
- Sep 20, 2026
Quick answer
A System One model is TypeSafe's name for an AI model that returns typed decisions and probabilities instead of generated text. Jev is TypeSafe's flagship model and the first System One model. The name refers to Kahneman's fast System 1 thinking.
A System One model makes fast, structured decisions that software can use directly. Jev is TypeSafe's flagship model and the first System One model.
This page explains the class in order to use Jev. Broader Decision AI benchmarks, datasets, and calibration studies belong on a future dedicated project, not on this hub.
How it differs from an LLM
| LLM | System One (Jev) | |
|---|---|---|
| Trained to | Produce text people read | Return calibrated decisions |
| Output | Tokens of prose or JSON-as-text | Typed answers + probabilities |
| Control flow | Often hidden in the prompt | Stays in your code |
| Confidence | Not the same contract | Choice/Score include confidence |
Official training name you will see in docs: RLCD (Reinforcement Learning for Calibrated Decisions). Calibration is across many predictions, not a promise about one ticket.
The decision primitives
You do not prompt Jev to “be a JSON API”. You pick a primitive:
Those three are how a System One model exposes an answer space.
A refund workflow
Official System One page walks a refund case:
- Build state: message, transactions, policy.
- Ask independent questions together.
- Combine answers with deterministic checks, then route.
Jev supplies the judgments. Code supplies eligibility math, idempotency, and the actual refund call.
Naming
TypeSafe chose “System One” for speed and focus, not because Jev is unconscious or approximate in the Kahneman sense. Do not stretch the metaphor into a product claim.
Jev's own name, per third-party reporting, nods at Jevons' paradox: cheaper decisions get used in more places. That is commentary, not an official SLA.
When to reach for a System One model
High-volume, low-latency gates in front of humans or LLMs: routing, filtering, scoring, tool selection.
When not to
If the user is waiting on a written answer, you still need a generator. If you need a ranked list of open-ended ideas, System One is the wrong class.
Common mistakes
- Treating “System One” as a second API beside Jev. The HTTP API is still
POST /v1/systemonewithmodel: "jev-latest". - Turning this site into a generic Decision AI platform. We stop at what you need to ship Jev.
Next: How Jev works or the API.
FAQ
Is System One a separate product from Jev?
No. System One is the model class. Jev is the first published model in that class.
Why 'System One'?
Official docs cite Daniel Kahneman's Thinking, Fast and Slow. System 1 is fast and intuitive; System 2 is slower and deliberate.
Sources
- System OneTypeSafe · accessed 2026-09-20 · documentation
- IntroductionTypeSafe · accessed 2026-09-20 · documentation