Skip to content
Jev Guide

Recipes

Six practical tasks, built with Jev.

Each recipe is one task, one primitive, and one working example. Explore it with sample input, or your own text when live testing is enabled, then read the result and copy the code.

  1. 01Route a messageChoiceClassify a customer message into the department that should handle it.
  2. 02Score urgencyScoreRate how urgently a message needs a response, on a three-level rubric.
  3. 03Check a replyNoulEstimate whether a draft reply addresses every explicit request in a customer message.
  4. 04Check a citationNoulEstimate whether a source passage actually supports the claim made about it.
  5. 05Rate a passageScoreRate how well a retrieved passage answers the question it was retrieved for.
  6. 06Pick a toolChoiceChoose which tool should handle a request, or none at all.

Every recipe shares the same server-side protection as the rest of Jev Guide: a fixed model and question, a per-visitor rate limit, and a shared daily cap. See how it works for the request and response shapes.

Not sure a bounded decision like these is the right fit for your use case? See Jev vs LLMs: when to use TypeSafe AI.