Pick a case or paste your own
8 recorded cases cover triage, moderation, routing, scoring, lead and candidate checks. Or start blank with your own text.
Jev AI playground · 20 welcome credits
Jev AI Playground
Try the Jev model on real text without writing code. Load one of 8 recorded cases or paste your own, ask typed questions, and get calibrated probabilities back in about half a second. Everything you run here can be exported as an API request.
- 1 credit per run
- ~0.5 s answers
- Yes/No · Choice · Score
- Failed runs refunded
Jev AI playground
3 Answers
Run Jev 1.14 to see answers
How the Jev playground works
Four steps from a blank text box to a request your application can send.
Write typed questions
A yes/no question returns a probability. A choice returns one winner and every option’s share. A score places the text on levels you define.
Run Jev and read the numbers
One run costs 1 credit and usually returns in under a second. Every answer carries a confidence you can branch on.
Export the exact request
Get API code renders your state, questions and model as cURL, JavaScript or Python. Save the questions as a judge to call them by ID.
The recorded cases
Each one was run on jev-1.13.0 and saved, so you can read the answers before spending a credit. Edit anything and run it again.
- Support ticket triageRoute a customer message to a team, flag urgency and read frustration in one call.
- Content moderationLabel a user comment, then decide whether a human moderator should look.
- Chatbot intent routingClassify what a user wants before your agent decides which tool or model to call.
- Product review scoringTurn free-text reviews into a star score and the topics your team tracks.
- Sales lead qualificationScore buying intent from an inbound form and pick the right follow-up.
- LLM guardrailScreen an incoming message for jailbreak and data-exfiltration attempts before it reaches your LLM.
- Citation checkCheck whether a quoted source actually supports the claim an LLM made.
- Candidate screeningRate a CV summary against a role with a rubric you control.
Need a template for a specific job? The use-case library has judges for LLM-as-a-judge, RAG evaluation, routing, entity matching, prompt-injection checks and more, each opening in this playground.
Models in the picker
Jev is the default. The same text and questions can be sent to the other hosted models for a second opinion.
| Model | What it is | Open with |
|---|---|---|
| Jev | TypeSafe’s System One model, jev-latest → jev-1.14.0. Text only through the alias, 64k context. Version guide | Run Jev |
| Jev 1.14 | The current release, of 14 October 2026, behind jev-latest and jev-preview; strongest gains in languages other than English. What changed | Run Jev 1.14 |
| Jev 1.13 | The previous release, pinned. Version guide | Run Jev 1.13 |
| Mercury Decide | Inception’s diffusion decision model on the same schema. Guide | Run Mercury Decide |
| Laya English · Multilingual | Convai’s small open decision models, short-text only. Guide | Run Laya |
| Clef · Clef Flash | Cloudflare’s Jev-compatible models, with image input. Guide | Run Clef |
After the playground
- 01
Create an API key
On the Jev API page. The key works against the TypeSafe-compatible endpoint at
/api/v1/systemone. - 02
Send the exported request
Paste the cURL, JavaScript or Python from “Get API code” into your service, or hand the setup prompt to your coding agent.
- 03
Batch what repeats
The batch page runs one judge over a CSV or a list of texts and returns a table.
Playground FAQ
Is the Jev AI playground free?
New accounts get 20 welcome credits and every playground run costs 1 credit, whatever the length of the text or the number of questions. Loading a saved example makes no live call. When the credits are used up you can buy a plan or a credit pack.
Do I need an account to try Jev?
You can read every recorded example without signing in. Running your own text, or an edited example, needs a free account so the result and the credits are stored with you.
Which model does the playground run?
Jev 1.14 (jev-1.14.0) by default, the release that jev-latest and jev-preview now point to, through a TypeSafe-compatible request; the picker also offers the jev-latest alias and the previous release, jev-1.13.0, pinned. The model picker also offers Laya English, Laya Multilingual, Mercury Decide and Clef, which are billed at their own rates.
Is this the official TypeSafe playground?
No. TypeSafe operates typesafe.ai and its own console. Jev AI is an independent playground and API with its own accounts, keys and balances. Choose the service you intend to use before creating an account.
Can the playground read images or audio?
Jev itself is text only. Choosing Clef in the model picker lets you attach images, because Clef accepts them; for Jev, convert the content to text first.
How do I turn a playground run into code?
Open “Get API code” under any result. It renders the same state, questions and model as cURL, JavaScript or Python against the Jev AI API, so what you tested is what you ship. Save the questions as a judge to reuse them by ID.
What does a failed run cost?
Nothing. If the upstream call fails, the credit is returned and the error is shown with its reason.