What is Jev?

In one sentence: give Jev application state and predefined questions; it returns typed answers with probabilities and confidence values that software can use.

How Jev works

TypeSafe AI describes Jev as a machine-native intelligence primitive. A developer provides unstructured state—such as a support message, transaction context or workflow record—along with questions whose answer shapes are known in advance.

The key idea is constraint. Jev gives up arbitrary string generation in exchange for output that fits a type declared by the application. This makes it suitable for places where software needs to classify, route, score, verify or branch.

Choice, Score and Noul

Choice

Selects from a predefined list. A support workflow might choose among billing, technical, sales or abuse, then return a probability distribution and confidence.

Score

Evaluates something against a rubric. Software can use the result as one signal in a larger workflow rather than treating it as an unquestionable verdict.

Noul

Represents the probability that a proposition is true. It is useful for yes/no-shaped checks where preserving uncertainty matters.

Is Jev an LLM?

TypeSafe AI positions Jev as a different model class rather than a smaller chat model. Both can accept unstructured information, but the output contract is different: a generative LLM produces strings token by token; Jev produces bounded, typed decisions in parallel.

Can Jev write text?

No—not in the open-ended sense expected from a chatbot or writing assistant. Its purpose is to choose, score and judge within structures defined by software. If your task needs an essay, conversational answer or newly written code, a generative model is the more natural tool.

Can Jev still be wrong?

Yes. A correctly typed output is not automatically a correct real-world judgment. Type safety prevents malformed answer shapes; probabilities and confidence help software manage uncertainty. Production systems still need evaluation, thresholds, monitoring and escalation paths.

Where to get Jev

Jev was announced in early access. Availability and integration details can change, so use the official TypeSafe AI website as the current source of truth.