Built for TypeSafe Jev

From business requirement to ready-to-run Jev questions

Describe what you want to automate. The tool finds the decisions hidden in your requirement, writes precise Choice, Score and Noul questions, tests them on Jev, and hands you validated JSON and code.

  • No sign-up
  • Bring your own key
  • Free model option
  • Nothing stored

Why this exists

Jev is only as good as the questions you ask it

Jev makes fast, focused judgments. Most bad results come from the questions it's given, not from the model.

Written by hand

  • One giant question. "Analyse this email" mixes routing, urgency and refunds into one judgment.
  • Overlapping options. "Billing" and "Payments" both fit the same message.
  • Vague scales. "1 = low, 3 = high" means something different to everyone.
  • Broken JSON. Found only when the API call fails.

Designed with this tool

  • Atomic questions. One judgment each, combined in your code.
  • Distinct options. Clear descriptions and a catch-all when needed. Overlaps are flagged.
  • Concrete levels. Each level describes a real situation, not a number.
  • Validated output. Checked against Jev's request format before you can export.

Three question types

Every decision fits one of three shapes

The tool picks the right type for each decision and tells you why. Jev answers each one with probabilities, not just a label.

Choice

Pick one from a list

For categories with no order: team, intent, document type, route.

Which team should handle this?

billing61%
technical35%
sales4%
Score

Place it on a scale

For ordered levels you describe: urgency, severity, quality, risk.

How severe is the issue?

Cosmetic

Workaround exists

Blocking

Score 1.43 of 2 · confidence 35%

Noul

Answer yes or no, with odds

For clear propositions where a probability lets you set your own threshold.

Does the customer ask for a refund?

93%probability of yes

How it works

Five steps, about five minutes

Start with step 1
  1. 1

    Set up your AI

    Free model or your own key. Paste, test, done.

  2. 2

    Describe the use case

    Plain words, plus examples, rules or documents if you have them.

  3. 3

    Answer a few questions

    Only when something would change the design.

  4. 4

    Review the questions

    See why each one exists. Edit anything.

  5. 5

    Test and export

    Run it on Jev, then copy JSON or code.

What you walk away with

Ready for your codebase, not a slide

  • Validated Jev JSONMatches the official request format. Export is blocked until it's valid.
  • Code in your languageTypeScript SDK, JavaScript, Python and cURL, generated from the same config.
  • A real test runRun it on a sample with your TypeSafe key and see the probabilities for every answer.
  • The reasoningWhy each question exists and what in your material it came from.
JSONTypeScriptPythoncURL
import { TypeSafeClient, choice, score, noul }
  from "@typesafe-ai/sdk";

const client = new TypeSafeClient();
const res = await client.systemOne({
  model: "jev-latest",
  state: email,
  questions: {
    department: choice("Which team should handle this?", { … }),
    urgency: score("How urgent is it?", [ … ]),
    refund_requested: noul("Does the customer ask for a refund?"),
  },
});

// res.answers.department.choice → "billing"

Use cases

Anywhere someone reads, then decides

Support, sales, finance, HR, legal, moderation, AI guardrails and more.

Explore all 20 use cases

Your keys, your data

Private by design

No accounts, no database, no logs of your content. Your keys are used for your request and nothing else.

1

An AI key

Designs your questions. A free OpenRouter key works (no card), or use Claude, OpenAI or Gemini. Only sent to that provider.

2

Your use case

A few sentences, plus examples, rules or Word, PDF and CSV files if you have them. Stays in your browser between steps.

3

A TypeSafe key (optional)

Only to run your questions on Jev. Held in the open tab and never saved.

FAQ

Common questions

Something else? The Jev docs cover the model itself.

Is it free?

Yes. No account and no payment. The design step uses your own AI key, and OpenRouter offers free models if you don't have one.

What is Jev?

Jev is TypeSafe's judgment model. You send it an input and a set of questions, and it answers each one with a label and probabilities, fast and consistently.

Do I need a TypeSafe key?

Only to run your configuration on Jev from inside the tool. You can design and export without it, and a simulated run shows you the results screen.

Is my data stored?

No. There is no database. Your text goes to the AI provider you choose for that one request. Your draft stays in your own browser, and you can clear it at any time.

Can I trust the AI's suggestions?

Treat them as a strong first draft. Each question shows why it was proposed and what to double-check, and you can edit everything before export.

Which AI models work best?

Larger models design better questions. Free models work for trying things out. For client work, use a capable paid model and avoid sending confidential material to free ones.

Design your first Jev configuration today

Free, no sign-up. Start with a free model or try the demo, and have your first draft in minutes.