Ten Levels of Jev

Checked, not fully reviewed. We ran its test suite — 201 pass, 10 live tests skipped without a key — and the whole demo against its mock backend. No Jev calls.

Thirty uses arranged in ten levels, three per level: single decisions as a smarter if, several picks in one call, composite scores with the weights in code, confidence-gated routing, intent and model routing, and on up to a coding agent that decides for itself when to ask. Each runs as a function you can read in one screen, and JEV_BACKEND=mock runs the lot offline with deterministic answers, which is how to read it the first time.

The repository is mostly a teaching artefact, and it is good at that: it puts the weights in code rather than the prompt, gates on confidence rather than trusting the top answer, and separates the decision from what the program does with it. Live mode talks to TypeSafe or OpenRouter at fixed addresses. There is a video walkthrough linked from the README.

MIT, 169 stars. Start here if the API reference reads as abstract.

See how it compares with other tools in Best Jev tools, tested hands-on.

Review updated Oct 4, 2026. Numbers quoted from the project are its author's own; we don't publish our own measurements of Jev.