Every task, in the right place.
Identify intent, assess priority, and route messages and tickets to the right handler.
Workflow routingMEET JEV · SYSTEM ONE MODEL
Meet Jev.
Built for decisions.
Intelligence. Into action.
Turn information into decisions. Put software in motion.
Meet Jev, a fast decision model built for automation.
THE DECISION ENGINEJEV / 01
Unstructured inputStructured decisions
70–500ms
End-to-end latency*
1 call
One call. Parallel answers.
Type-safe
Typed values, ready for code
P(decision)
A probability for every decision
* Latency source: TypeSafe AI announcement. Actual performance varies by task and operating conditions.
01 / A DIFFERENT KIND OF INTELLIGENCE
What is Jev?
Jev is TypeSafe AI’s first public System One model, built for automated decisions inside software.
Provide a state and predefined questions. Jev returns typed answers and their probabilities so your software can choose the next step using business rules.
Try the decision flow below02 / FROM STATE TO DECISION
How does Jev work?
Provide context, define question types, then read structured decisions. The preset examples below show support routing, content review, and lead scoring.
01 / STATEInput state
TYPED QUESTIONSPredefined questions
intentenumneeds_humanbooleanurgencyscore 1–5
JEV
02 / DECISIONStructured output
Example result
{
"intent": {
"value": "refund",
"probability": 0.96
},
"needs_human": {
"value": true,
"probability": 0.92
},
"urgency": {
"value": 4,
"confidence": 0.89
}
}Example next step: route to a billing specialist for priority review of the duplicate charge.
01
Provide context
Messages, logs, or business events,
just as they are.
02
Define the answer's shape
Categories, booleans, or scores:
you choose the output type.
03
Let software act on decisions
Read the results and probabilities,
then apply business thresholds to choose the next step.
03 / A NEW PRIMITIVE
How does Jev differ from LLMs?
Jev returns typed decisions and probabilities for predefined questions. LLMs focus on general generation and reasoning, and can also constrain JSON through structured outputs.
Type safety ≠ a correct decision.
Critical workflows still need thresholds, validation, and human oversight.
| Dimension | Language models | Jev · System One |
|---|---|---|
| Output | Text or constrained JSON | Values of predefined types |
| Process | Token-by-token generation | Multiple decisions in parallel |
| Integration | Parsing or Schema constraints | Read structured results |
| Probability | Depends on the model and API | Returns calibrated probabilities |
Based on Jev's published positioning. Actual capabilities depend on the API. TypeSafe AI · Jev official website
04 / INTELLIGENCE, EVERYWHERE
What can you use Jev for?
Jev can support workflow routing, real-time interaction, batch feature extraction, and model safety assessment.
Feedback that keeps up.
Assess the current state quickly to inform decisions in latency-sensitive interactive systems.
Real-time interactionFind signals in the noise.
Turn unstructured content into categories and scores that software can process.
Batch feature extractionOne more check before acting.
Assess risks in prompts and model outputs to help systems decide whether to allow, review, or block them.
Model safety assessment05 / LIVE SIGNAL
Jev updates
From the official blog, docs, SDK, X, and public coverage. Items stay in their original language.
Filter sources
Checked . Original language; public links and titles only.
awesome-typesafe: a community field guide for Jev
Independent collection of SDKs, demos, agent tools and evals. Not affiliated with TypeSafe AI.
TypeSafe Python SDK v0.7.1
Early key checks omit secrets from exception logs; gateway samples added.
Jev for Devs: typed questions, parallel answers, priced on input tokens
Public developer thread on Choice / Score / Noul, latency, and $0.042 per million input tokens.
TypeSafe AI Releases Jev: A System One Model That Returns Typed, Calibrated Decisions Instead of Text
Asif Razzaq covers Choice, Score and Noul, hosted-API waitlist, and flags vendor-run speed/cost demos.
TypeSafe Python SDK v0.7.0 — pydantic response models
Breaking change: ser/de moves from msgspec to pydantic. system_one can take a pydantic response_model.
LangChain on Jev as an agent evaluator
Public post comparing Jev with GPT/Claude for evaluation consistency, cost and latency.
TypeSafe AI exits stealth with $40M to build AI for use by software
Paul Gillin reports a $40 million seed led by DCVC and Jev as typed judgments for applications, not chat.
Introducing System One Models and Jev
Launch thread on the official blog post. Discussion centers on speed claims and what “no hallucination” means for typed outputs.
Vercel AI SDK: evaluate with Jev
Public note on wiring Jev into the AI SDK evaluate path for structured checks.
Introducing System One Models & Jev
Official launch of System One and Jev: typed decisions, parallel answers, RLCD, and 70–500ms end-to-end response times as vendor figures.
TypeSafe Python SDK v0.5.7 — first public release
Initial public Python SDK for calling System One / Jev.
Models: jev-1.13 / jev-latest
Current public model identifiers, question types and API contract. Product facts belong here, not on this introduction site.
System One: state in, typed decisions out
Canonical explanation of System One: unstructured state plus typed questions, calibrated probabilities, no free-text generation.
AI: too good to be true, too bad to be useful
Pre-launch essay on why impressive AI still fails as a software primitive — the argument that precedes System One.
No items for this source.
LESS WAITING. MORE DOING.
Make intelligence part of the action.
Start with a question. Explore the next decision Jev could make for your system.
Get early API accessRequest access at typesafe.ai