Jev: route, check and score with a structured decision model

A decision your application can use

Jev returns structured judgments rather than a conversational answer. A request provides a state—the relevant information—and questions about it. Your application still decides what to do with the returned values.

The tutorial uses fictional examples: routing a customer inquiry, checking a writing brief and rating a bug report. These demonstrations explain the request pattern; they do not establish accuracy across a real production workload.

Choose a question type

The three text question types covered are Choice, Noul and Score. Use the video’s examples to see how selecting an option, checking a condition and assigning a score differ.

1. Prepare the plain-text state containing the facts the decision needs.
2. Choose a question type and write a specific question.
3. Define the options or rules the model should use.
4. Inspect the returned decision and any probability or confidence information.

Run a small Playground check

The walkthrough uses the official Playground and jev-latest. At recording, running requests required account credit and the Add funds dialog requested a five-dollar minimum. Check the current interface and price before spending.

TypeSafe’s listed price in the video was $0.042 per million input tokens with output tokens free. A low per-token price is not the same as a free account or evidence that an automated action is safe.

1. Check account credit and the current price.
2. Enter a small fictional state in the Playground.
3. Run one routing, one brief-checking or one severity-scoring example.
4. Compare the result with your own expected answer before wiring it to an action.

The takeaway

Use structured judgments as inputs to your code, and evaluate them on examples with known answers before automating consequences.