
TypeSafe AI's System One model, built...
Split a large judgment into small questions that code can branch on, then send those to Jev. Choice fits unordered options such as a department or a document type. Score fits ordered judgments such as severity or urgency. Noul fits yes-or-no questions such as whether a refund was requested or whether personal data is present. Questions in one request share a state, stay independent, and return together. If a later question cannot be built until the first answer arrives, the docs say to send a second request. When confidence is low, hand the case to a person or a reasoning model instead of acting immediately. It fits backend developers working on ticket routing, content review, and retrieval filtering.
Jev is TypeSafe AI's System One model, built for fast decisions that software can branch on. A call sends a state, which may be a string, a JSON object, or an array of text, plus questions you write. It does not generate a reply, code, or an explanation of its reasoning. It returns a typed result inside the range you set. Choice picks one of your options and returns a probability for each. Score places the input on levels you define. Noul returns the probability that a statement is true. Choice and Score also include confidence, so code can act automatically or hand the case to a person. Input is text only for now. Images, audio, and video are not supported. Official client libraries are available, and the HTTP API can take several questions in one call.
Difficulty: Intermediate
Choice decisions
Picks one of the options you supply and returns a probability for every option plus confidence for that judgment.
Score ratings
Places the input on ordered levels you write. The score can fall between two levels and includes a legend and probabilities.
Noul yes-or-no probability
Returns the probability that a yes-or-no statement is true. The docs say this answer has no separate confidence field.
Several questions per call
Several questions read the same state, are judged independently, and come back in one response under the ids you chose.
Route support tickets
Put the customer message and the refund policy in the state, use Choice for the request type, and Noul for whether a refund was asked.
Screen messages before they proceed
Ask whether text entering or leaving an app contains personal data or crosses a line, then let code pass, review, or block from the probability.
Filter retrieved passages
Score each retrieved passage, then let code decide which passages may continue into the answering step.
The official docs say it does not generate a reply, code, or an explanation of its reasoning. You define the questions and the allowed answers, and it returns typed results and probabilities. If the product still needs a paragraph for a user, code should take the Choice, Score, or Noul result and call another step. This call is not the step that writes the message.
The docs suggest putting questions that share one state into the same request. Those questions are judged independently and return together, and the types can be mixed. A second request is for the case where the next question cannot be built until the first answer exists. You choose the question ids and use them to read each answer.
Real reviews and feedback from users