How urgent? Which team?Let your rule decide.
Write down, in your words, what counts as urgent and where each request belongs. See how often that holds on your own past cases. Then engineering gets the rule, with tests, and wires it in.
Every team has rules nobody wrote down
Which request is urgent, which review needs a reply, which lead deserves a sales call. Today those calls live in people’s heads, differ from person to person, and take weeks to reach the product. Jev Studio lets the person who knows the rule write it down in plain words, see how often it is right, and hand it over finished.
What is Jev?
Jev is a decision model by TypeSafe AI. It reads a text and answers only with your options: yes or no, one choice, or a place on a scale, each with a number for how sure it is. It writes no text.
TypeSafe documentationHow it works
Four steps from a hunch to a checked rule.
- 01
Ask in your words
Write the question the way you would ask a colleague: Is this urgent? Which team should take it? How upset is this customer?
- 02
See the answer and how sure it is
Every answer comes with a number. Change a word in the question and watch the number move, so you see which wording is clearer.
- 03
Check it on real cases
Add requests you have seen, with the answer each should get. Jev Studio shows how many go the right way, which ones do not, and where to draw the line between yes and no.
- 04
Hand it over
Engineering gets the finished rule: the request to send, code with your line, and your cases as tests. Nothing to translate, no meeting to explain it.
Watch it decide
Three example texts, three questions each. Pick one or let it play.
We waited forty minutes for a table we had booked, and the risotto arrived lukewarm. That said, our waiter was genuinely lovely and comped the dessert without us asking. Not sure we will be back.
375 ms for all three questions
Should the owner answer this review personally within a day?
Yes or no0.82Yes(line at 0.60)What is the overall sentiment of this review?
Pick one0.88mixed- mixed 0.88
- negative 0.12
- positive 0.00
How many stars would the author most likely give?
Scale2 stars1.26 / 4
Why not just ask a chatbot?
An AI chat can label a text too. The difference is what you can check and hand over.
| Aspect | Jev Studio | An AI chat |
|---|---|---|
| What comes back | Jev Studio:An answer from your options, with a number for how sure it is | An AI chat:A sentence someone has to read or parse |
| Possible answers | Jev Studio:Only the options you set up | An AI chat:Anything, including options you never offered |
| Where the rule lives | Jev Studio:A visible line you set and test on your cases | An AI chat:Hidden in the wording of a prompt |
| Cost | Jev Studio:About 2 cents per 1,000 short texts with three questions (list price, September 2026) | An AI chat:You pay for reading and writing; cost varies with the length of the answer |
| Best for | Jev Studio:Sorting, routing, flagging, prioritizing | An AI chat:Writing replies, summaries, open questions |
What we have measured so far
- for three questions about one text, median of 20 calls to the Jev API
- 0.23 s
- to check five labeled cases, four calls at a time, as Jev Studio does
- 0.43 s
- test descriptions the question assistant sent the right way: real Jev cases to a draft, writing jobs to a language model
- 16 of 16
- real Jev calls in the last 30 days: 99.6 % answered, 95 % of them in under 0.6 s
- 50,000+
- per 1,000 short texts with three questions, at TypeSafe’s list price
- about 2¢
Questions
Do I need to write code?
No. You write questions in plain words and add example cases. Code only appears when you hand the finished rule to engineering, and they get it ready to use.
Which decisions does Jev Studio fit?
Decisions about a piece of text where the possible answers are known in advance: urgent or not, which team, which category, how satisfied on a scale. If you need text written, such as replies or summaries, a language model is the better tool, and Jev Studio says so.
How do I know the rule is right often enough?
Add cases with the answer each should get and check them. Jev Studio shows how many go the right way, lists the ones that do not, and for yes-or-no questions suggests the line that gets the most cases right with the widest margin.
What does engineering get?
The finished call to TypeSafe’s API, TypeScript that turns each answer into your decision with your line, and your labeled cases as tests. Whoever owns the technical side sees cost per call and the exact interface with its types before anything runs.
What does it cost to run?
About 2 cents per 1,000 short texts with three questions (TypeSafe’s list price, September 2026). Only the text Jev reads is billed: $0.042 per million input tokens. Jev Studio adds a daily limit per person.
Where does our data go?
Your rules and cases are stored on a private server in Germany. The texts and questions of each check are sent to TypeSafe, a US processor, to be answered. Only add cases you are allowed to share with such a processor. Descriptions you give the question assistant inside the studio also go to Cursor, which drafts the questions.
Who can use Jev Studio?
Anyone on the waitlist, once invited. Leave your email address below; when it is your turn, you get a one-time link that works for seven days. Each person sees only their own rules and cases.
Join the waitlist.
Jev Studio opens a few places at a time. Leave your address and, if you like, the decision you would write down first; when it is your turn, we send you an invitation link by email.
Already invited? Sign in