Start here / JevModel
How to use Jev: a step-by-step guide
Learn how to use the Jev model: try it in the playground, write Choice, Score and Noul questions, read probabilities, then call the Jev API and run batches.
Last updated:
At a glance
- Jev answers bounded questions about a state. Give it the evidence, name the possible answers, and let your code act on the label, score or probability it returns.
- Start in the playground with one live run and no signup. Verify an email for 5 free runs a day, then move the same request to the API.
1. Run your first decision in the playground
Open the Jev playground and pick the support routing template. Replace the sample with a short message of your own, such as “I was charged twice this month”, and press Decide. The live panel shows the chosen queue and its probability. This first run needs no account and shows the full loop: a state goes in and a typed answer comes back.
Your first Jev decision2. Write the state: evidence, not instructions
The state is what Jev judges. It can be plain text, a JSON object or a JSON array, up to 8,000 characters. Include only facts that affect the decision, such as the message, the customer plan or the order status. Do not hint at the answer inside the state, and remove secrets or personal data you do not need. A JSON object with clear field names is often easier for the model to use than a long paragraph.
Choice, Score, and Noul{
"message": "I was charged twice this month and need a refund.",
"plan": "Pro",
"open_tickets": 0
}3. Pick the question type that matches the decision
Use Choice when the answer is one of several named options, such as which team handles a ticket. Use Score when the answer is a level on an ordered scale, such as urgency from 1 to 4; write one criterion per level. Use Noul for a yes-or-no proposition, such as “a human should review this now”, and read the probability that it is true. Each request can ask 1 to 8 questions about the same state, and they run together.
Jev prompt and question examples4. Read probabilities and set thresholds
Jev returns probabilities, not guarantees. A Choice of billing with probability 0.92 means the model strongly prefers billing under your option definitions; it does not mean 92% accuracy. Decide in code what happens at each level: act automatically above a high threshold, send borderline cases to a person, and log everything. Tune thresholds on a set of your own labeled examples, including ambiguous ones.
Jev probability, confidence and human-review thresholdsEvaluate a Jev workflow5. Call the Jev API and scale up
After email verification, create an API key in the Console and POST the same state and questions to /api/v1/systemone from your server. The first 5 successful runs each UTC day are free; after that, requests use input tokens from your balance. To classify a file, upload CSV, TXT or JSONL to Batch. To chain questions, save a Decision Flow and run it with one request.
JevModel API: request and responseBatch text classification with the JevModel APIAI Decision Flows with Jev and human reviewcurl -X POST https://jevmodel.app/api/v1/systemone \
-H "Authorization: Bearer $JEVMODEL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"state":"I was charged twice this month.","questions":{"review":{"type":"noul","instructions":"Should a billing specialist review this now?","criteria":{"true":"A specialist should review it now.","false":"No immediate review is needed."}}}}'Common mistakes to avoid
Overlapping options make Choice answers unstable, so give each option a distinct definition and add an “other” or “needs review” option. Asking Jev to write text will fail; it only returns labels, scores and probabilities. Validation errors usually come from question IDs with capital letters, instructions shorter than 8 characters, or a Choice with fewer than 2 options. Do not treat one good run as proof; test clear, mixed and adversarial cases.
Jev vs GPT and Claude: choosing a decision layerFrequently asked questions
Do I need to code to use Jev?
No. You can run decisions in the playground and classify files in Batch without code. You need code only to call the API from your own application.
Is Jev free to use?
On JevModel you get one live run per day without signup and 5 free successful runs per UTC day after email verification. Additional runs use a token balance.
Can Jev generate text or chat?
No. Jev is a decision model. It returns typed answers to questions you define. Use a text-generation model for replies and summaries.
JevModel is independent and not affiliated with TypeSafe AI.