Agent tooling

jev-code

@FrancoisChastel★ 1TypeScriptMITupdated 2026-09-19▶ runnable

Jev, TypeSafe's System One classifier, as a tool inside Claude Code, Codex, Pi, and OpenCode: typed classify, check, score, rank, and ask, plus one-command setup.

FrancoisChastel/jev-code ↗

Where it calls Jev

TYPESAFE_BASE_URL       Default https://api.typesafe.ai

src/cli/help.ts:40

The link points at the commit we read, so the line number still holds.

What it asks Jev

These question sets are lifted from this project's source exactly as written.

from src/cli/doctor.ts:102

  1. alivetrue/false

    Is this a single short word?

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    alive: { type: 'boolean', instructions: 'Is this a single short word?' },
  },
});

from tests/core/client.test.ts:8

  1. qtrue/false

    Is this a greeting?

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    q: { type: 'boolean', instructions: 'Is this a greeting?' },
  },
});

Other projects in this category