Jev with Claude Code: setup and routing tutorial

This tutorial explains how the speaker connects Jev to Claude Code and uses it to make classification decisions within an AI agent workflow. He describes Jev as a classifier rather than a large language model: its responses are limited to true or false, a choice from supplied options, or a score on a scale. The setup uses an API key obtained directly from the provider or access through OpenRouter, which he used for his examples. The tutorial does not demonstrate how to run models locally.

The first examples cover model routing and skill selection. Jev chooses a model for each task or selects a skill for Claude to load. In the speaker's 12-prompt routing test, he reports 70% savings compared with his fixed-model baseline; its name is unclear in the captions. He advises checking whether the resulting output meets the needs of the work and suggests a command to switch routing on or off.

A separate demonstration compares Jev, Haiku and Fable on classifying 100 emails as leads. The speaker reports that Jev finishes in under a second; this is a result from his demonstration, not a general performance guarantee. Later examples include searching images and videos by meaning in Rubric and an overview of Unclutter, a Chrome extension that he says uses Jev to identify and remove ads and cookie banners.