Getting Started with Jev

You can be useful with Jev in one afternoon: one key, one request, one branch in code.

1. Get access and a key

TypeSafe's official path is the dashboard after you have access. The 2026-09-15 launch post describes early access and a waitlist at typesafe.ai. We have not independently timed waitlist access.

If you already bill through a gateway, you can call the same model as Vercel Jev (typesafe-ai/jev) or OpenRouter Jev (~typesafe/jev-latest). Details: Jev on OpenRouter and Vercel. The longer “how to use Jev” path is here.

Set the key in your environment. Never commit it.

export TYPESAFE_API_KEY=your_key_here

2. Try the official Playground

Official quickstart: open the Playground, paste a state, add a Noul such as “Does this message express urgency?”, then mix in Choice and Score. Use that to learn question wording before you write production code.

This site's own Playground is not shipped yet. We will not pretend a local widget is a live Jev call.

3. One cURL request

Endpoint from official API docs:

POST https://api.typesafe.ai/v1/systemone
Authorization: Bearer <API_KEY>
Content-Type: application/json

curl -X POST https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Hi, I have been trying to connect my Stripe account for 3 days and the integration keeps failing. I am losing sales. Please help ASAP.",
    "model": "jev-latest",
    "questions": {
      "urgency": {
        "type": "noul",
        "instructions": "Does this message express urgency?"
      }
    }
  }'

Expected shape — values are example output, copied from TypeSafe's published quickstart, not a measurement from this site:

{
  "model": "jev-1.13.0",
  "answers": {
    "urgency": { "type": "noul", "noul": 1.0 }
  },
  "usage": { "input_tokens": 392, "output_tokens": 65 }
}

More HTTP detail lives in the cURL tutorial and the API page.

4. Python

Official SDK package: typesafe-sdk. Python 3.10+ according to the quickstart.

pip install typesafe-sdk

from typesafe_sdk import Noul, TypeSafeClient

with TypeSafeClient() as client:
    response = client.system_one(
        state="The export button crashes Settings in Safari.",
        questions={
            "is_bug": Noul(instructions="Does this message report something broken?"),
        },
    )

print(response.answers["is_bug"].noul)

Full walkthrough: Jev Python.

5. TypeScript

Official package: @typesafe-ai/sdk. Node.js 20+.

npm install @typesafe-ai/sdk

import { noul, TypeSafeClient } from "@typesafe-ai/sdk";

const client = new TypeSafeClient();
const response = await client.systemOne({
  state: "The export button crashes Settings in Safari.",
  questions: {
    is_bug: noul("Does this message report something broken?"),
  },
});

console.log(response.answers.is_bug.noul);

Full walkthrough: Jev TypeScript.

6. Put a decision in code

The model should not “do the refund”. Your code should:

if (response.answers.is_bug.noul > 0.8) {
  assignTo("engineering");
} else {
  assignTo("triage");
}

Pick thresholds after you look at your own tickets. Official confidence guidance is a starting pattern, not a shipped default.

When to use this path

Use Getting Started when you need a working request today. Expand to mixed questions once one Noul looks sane.

When not to start here

If you still need to decide whether Jev is the right tool, read What is Jev and limitations first.

Common mistakes

• Forgetting Authorization: Bearer.
• Putting the English question only in the key (is_urgent) and leaving instructions vague. Official docs say IDs are not sent to the model.
• Shipping the first threshold that “feels right” on five examples.

Check pricing before you load-test, and skim examples for request shapes that match real products.