AbdelStark/awesome-typesafe-jev

Awesome Jev is a curated, source-backed field guide to TypeSafe’s System One models, built by AbdelStark and published as both a GitHub repo and a browsable site. Its opening table is organized around what someone wants to do: understand the idea in a couple of minutes, make a first typed call, see live projects, or test TypeSafe’s own claims against independent evaluations.

The list documents one worked example directly: TypeSafe’s published support-ticket sample, where a single Jev call returns three typed answers at once. Given a customer message about a failed Stripe integration, Jev’s Choice answer picks “technical” at 0.85 probability, its Score answer rates frustration at 1 on a 0-2 scale (“frustrated but civil”), and its Noul answer, a probability for a single yes/no question, rates urgency at 1.0. The README is careful to label this a documented example rather than a live call, and stresses that application code still decides what to do with the answer, whether that means routing a ticket or escalating it.

Beyond that walkthrough, the repo curates sections for official docs and SDKs, agent tooling, browser agents, games and robotics, evaluations and independent research, and showcases, including entries for other projects like pg-jev and open Jev-style models with released weights.

How it uses Jev

  : mode === 'noul'
    ? 'console.log(answers.decision.noul);'
    : 'console.log(answers.decision.score, answers.decision.legend);';
code = [
  `import { ${mode}, TypeSafeClient } from '@typesafe-ai/sdk';`,
  '',
  'const { answers } = await new TypeSafeClient().systemOne({',
  `  state: { text: ${quoted(state)} },`,
  '  questions: {',
  answer,
  '  },',
  '});',
  '',
  output,
  '',

View in assets/designer.js

Open on GitHub