01Guide
What is Jev? A decision model, not another chatbot
A plain-language explanation of Jev by TypeSafe AI and of the difference between a model’s decision and the execution of an action.
Open guideDECISION AI, EXPLAINED.
Fewer words.
More decisions.
Jev is an AI model that returns decisions instead of paragraphs. Understand what it can do, calculate the costs and prepare your first integration.
An independent guide to TypeSafe AI’s model. We keep facts separate from marketing. Source of the explanation: TypeSafe.
THE DECISION PRINCIPLE
INPUT / SAMPLE MESSAGE
“The package arrived damaged.Please send a replacement.”
↓ bounded questions ↓
CHOICEWhich category?
NOULRequesting a replacement?
SCOREIs the information sufficient?
DIAGRAM, NOT A LIVE CALL · Code controls actions and permissions.
0.042 USD
Per million input tokens.
S01 · as of 23 September 2026
Not another chatbot
Bounded answers for software.
S02
START HERE
From “what is it?” to first use.
No big promises. With explanations, examples and links to the original sources.
02Guide
Choice, Score and Noul: three question types
How to build a Choice, Score or Noul question, what the answers mean, and why a score is not a precisely measured quantity.
Open guide03Use cases
Jev for e-commerce: from categories to customer support
Proposed pilots for online shop catalogs, product search and customer inquiries, with no false promises about automation.
Open guideLATEST ARTICLES
What’s new with Jev.
News, how-tos and research. A source for every claim.
18 October 2026Research
Jev as a dispatcher for multiple agents: fewer tokens, similar accuracy
The Jev-LiteMAS preprint let Jev route agents and screen intermediate results. Generative-model tokens fell by 12.7 to 54.6%, while accuracy stayed close.
Read article17 October 2026Research
Jev as a RAG answer checker: similar accuracy, more errors let through
A preprint compared Jev 1.13 with GPT-6 Astra on RAG claim verification. Accuracy was similar and Jev far cheaper, but it passed unsupported claims more often.
Read article16 October 2026Research
One extra option shifts Jev's probabilities: what a new preprint found
A preprint from 12 October 2026 reports that jev-1.13.0 probabilities depend on which other options a Choice question lists. What it means for online shops.
Read articleTRY THE PRINCIPLE
Not just reading. Tools, too.
They run in your browser. Do not enter sensitive data; none of the tools here calls Jev.
01Local calculation
Cost calculator
Enter your call volume and input length. You get a transparent estimate in USD.
Open tool02No API call
API question builder
Assemble a Choice, Score or Noul question and export the request body.
Open tool03Labeled simulation
Decision lab
Move the threshold across hand-made examples. You will also see a confident mistake.
Open toolThe right format
is not the right answer.
A model can return an allowed category and still be wrong. That is why we explain limits and testing before we start talking about automation.
WHAT WE CARE ABOUT
A source for every claim.
A date for every price. Limits for every test.
Vendor documentation, another author’s measurement and our own proposal are not the same thing. The register also lists sources whose full content could not be retrieved.
No fictitious reviews, no promises of flawlessness and no passing off a simulation as live AI.
How we work with sources ↗QUESTIONS BEFORE YOUR FIRST TEST
The essentials in one place.
Is JEV.sk the model’s official website?
No. It is an independent Slovak guide. Jev is made by TypeSafe AI.
Can I try the real model here?
No. The calculator does arithmetic, the builder creates JSON and the simulator uses hand-prepared values. A live API would require a separate server-side integration.
Can I pass customer emails to the model?
Before a live pilot, you need to review the data flow, contracts, data minimization and permissions. To start, we suggest public product data. More on security.
Why don’t you say that Jev is always the best?
The right choice depends on the task, the errors, the costs and the oversight. That is why we are preparing a methodology for a Slovak-language test instead of a general leaderboard. How to test.