Your First Jev Call: A 5-Minute Quickstart

This walkthrough uses TypeSafe’s own API. Every snippet below was run against jev-1.13.0 on September 19, 2026. If you are calling Jev through OpenRouter, Vercel or Cloudflare instead, the idea is the same but the endpoint and model name change; see Jev channels compared.

Before you start

You need a TypeSafe account that has been let in from the waitlist, and an API key from the console. If you do not have one yet, see How to get a Jev API key. Then put the key in your environment:

export TYPESAFE_API_KEY="paste-your-key-here"

Step 1: Ask one yes/no question

A Jev request has three parts: the model, the state (the text you want judged) and questions. Start with a single Noul, Jev’s yes/no question type.

curl

curl -s https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-1.13.0",
    "state": "Hi, can you help me reset my password? The reset email never arrives.",
    "questions": {
      "needs_access_help": {
        "type": "noul",
        "instructions": "Is the user asking for help getting into their account?"
      }
    }
  }'

Python

# pip install typesafe-sdk
from typesafe_sdk import Noul, TypeSafeClient

client = TypeSafeClient(model="jev-1.13.0")

response = client.system_one(
    state="Hi, can you help me reset my password? The reset email never arrives.",
    questions={
        "needs_access_help": Noul(
            instructions="Is the user asking for help getting into their account?"
        )
    },
)
print(response.answers["needs_access_help"].noul)

JavaScript

// npm install @typesafe-ai/sdk
import { TypeSafeClient, noul } from '@typesafe-ai/sdk';

const client = new TypeSafeClient({ defaultModel: 'jev-1.13.0' });

const response = await client.systemOne({
  state: 'Hi, can you help me reset my password? The reset email never arrives.',
  questions: {
    needsAccessHelp: noul('Is the user asking for help getting into their account?'),
  },
});
console.log(response.answers.needsAccessHelp.noul);

Step 2: Read the response

With curl you get the raw JSON. It has this shape:

{
  "model": "jev-1.13.0",
  "answers": {
    "needs_access_help": { "type": "noul", "noul": 0.99 }
  },
  "usage": { "input_tokens": 287, "output_tokens": 20 }
}

• answers holds one entry per question, under the ID you chose.
• noul is the probability that the answer is yes. 0.99 is a confident yes.
• usage.input_tokens is what you pay for: at $0.042 per million tokens, this call cost about $0.000012. Output tokens are free.
• model shows the exact version that answered.

Step 3: Ask several questions at once

Jev answers every question in a request in parallel, so adding questions barely changes the response time. Mix the three types:

Request body

{
  "model": "jev-1.13.0",
  "state": "My invoice lists two seats, but only one of us can sign in, and the login page keeps timing out.",
  "questions": {
    "is_bug": { "type": "noul", "instructions": "Does the customer describe something not working?" },
    "queue": {
      "type": "choice",
      "instructions": "Which team should handle this message?",
      "criteria": {
        "billing": "Charges, invoices, seats on the bill",
        "technical": "Bugs, outages, login problems",
        "other": null
      }
    },
    "urgency": {
      "type": "score",
      "instructions": "How urgent is this message?",
      "criteria": ["Can wait a week", "This week", "Today"]
    }
  }
}

In our run, Jev returned is_bug 0.98, queue = technical with confidence 0.94, and urgency 1.83 on the 0 to 2 scale, meaning close to “Today”. A Choice returns the winning option plus a probability for every option; a Score returns a position that can fall between levels. See Choice and Score.

Step 4: Use the answer in code

The point of Jev is that the answers go straight into logic, with no parsing:

answers = response.answers
if answers["queue"].confidence < 0.6:
    send_to_human(ticket)
elif answers["queue"].choice == "technical" and answers["urgency"].score > 1.5:
    page_on_call_engineer(ticket)
else:
    assign(ticket, team=answers["queue"].choice)

Thresholds like 0.6 and 1.5 are yours to choose. Start conservative and adjust after checking real examples; see Confidence.

If something goes wrong

You see	Fix
401 “Cannot authenticate with the server”	Check the key. Details
403 “Must supply an API key!”	The Authorization header is missing. Details
400 “Unknown model: jev-1.13”	Use jev-1.13.0 or jev-latest. Details
422 “Field required”	A field such as model or a Choice’s criteria is missing. Details

Next steps

• SDK details: Python and JavaScript
• How to write better questions: Noul, Choice, Score, State
• What it costs at scale: Jev pricing