Jev, for people who build things

A six-lesson field course on TypeSafe AI's decision model. About three hours in total, in 25-minute pieces. No API key needed to follow along.

What this is

Jev is not a chat model. You give it text and a set of questions with options you define, and it returns calibrated probabilities over those options in about 100 milliseconds, for a few thousandths of a cent. That makes it a very good fit for the decisions inside voice agents, n8n and Make flows, CRM webhooks and support inboxes: which team, how angry, call back or not, act or ask a human.

This course grew out of a community benchmark of 40 real German sales calls (39 right, one confidently wrong). Every claim links to the TypeSafe docs, every exercise opens in the TypeSafe Playground on your own account, and the "simulated playground" widgets replay real responses recorded on 2026-09-19 so you can see actual numbers before spending your own credits.

Lessons

Reference

How to use it

In order, one sitting eachEach lesson is a ladder of six short levels with a quiz at the end. Arrow keys move between levels. Carry one real decision from your own work through all six lessons.
Playground linksEvery request block has an "open in Playground" link. It pre-fills state and questions on console.typesafe.ai; you press Run. New accounts come with free credits.
Replay widgetsBlocks marked "simulated playground" replay responses recorded from jev-1.13.0 on 2026-09-19, with tokens, cost and timing. They are real data, not mock-ups.
Bring back what you findRun the lesson 6 shadow evaluation on your own traffic and post the band table in the community thread or on the TypeSafe Discord.

Primary sources, in reading order

  1. docs.typesafe.ai/primitives: the three primitives and the request shape.
  2. docs.typesafe.ai/confidence: confidence and thresholds.
  3. Model jaggedness, jev-1.13: what breaks, per version.
  4. docs.typesafe.ai/patterns: fan-out, routing, composite scoring.
  5. The community benchmark brief (2026-09-17).
  6. TypeSafe AI vs LLMs in Wikipedia Race (video, 4 min).
Course index

Written for the Flow Dynamics AI community, September 2026. Facts verified against docs.typesafe.ai on 2026-09-19 for jev-1.13.0. Not affiliated with TypeSafe AI. Questions: TypeSafe Discord or the community thread.