Try System One models on jevtypesafe.net with 10,000 signup credits. Classify text, test LLM guardrails, route intents or score inputs — with Jev and Laya models.
Start with the filled-in ticket or select a scenario above. Choose between Jev & Laya models and run live on jevtypesafe.net.
Choice · Yes/No · Score — one question per run
A support ticket needs a destination. With Choice, you define the possible answers and Jev or Laya returns a judgment across those options. The result gives your code a named answer to work with instead of an essay to parse.
Read the probabilities alongside the selected answer. They show how the model weighs the available answers for this question and input. They do not certify that a ticket belongs to a particular team.
Keep the raw response when you want to inspect exactly what was returned. System One models evaluate text; they do not write a reply to the customer or explain their reasoning. You decide what your application does with the judgment.
Try System One models online with the same message and three different questions. You can see what changes when you ask for a category, a yes/no judgment, or a score without learning a new scenario each time.
Which team should handle this support ticket?
Ask which team should handle the message. Use Billing, Technical, and Account as the possible destinations, with a short definition for each. This is useful when your next step depends on a specific category.
Does this message describe a problem that needs attention today?
A deadline can matter even when the message sounds polite. Ask about urgency separately from the team assignment so your code can treat the two decisions differently. The probability describes the model's judgment, not a firm rule.
How frustrated does the customer sound?
Use a defined scale when a degree is more useful than a category. Read the score against those definitions, rather than treating it as an objective measurement of the customer.
Each example asks a separate question about the same fictional ticket. Output formats shown for illustration.
Follow a fictional support ticket from input to judgment. Each published example includes its question and the actual response from a JevTypeSafe run, so you can inspect what the model returned.
Run the ticket above to inspect your own result. Published examples here pair a fictional input with its question and a recorded response.
No recorded judgments are available yet. Your inputs and responses are not published by default.
The ticket is a starting point. Try System One models with a task where you already know the answers your software can act on. Define the question and categories in terms that make sense for your workflow.
Your feedback list mixes bug reports, feature requests, and general comments. Ask which kind of feedback a message contains, with a definition for each category. Your code can use the answer to suggest an initial status or assignee.
You have a reading list and a specific topic in mind. Ask how relevant an article is to that topic on a defined scale. Use the judgment to help prioritize what you read, then inspect the source before relying on it.
Your application receives requests that belong in different flows. Define the handlers as answer options and ask which one fits the input. Keep the model's classification separate from your code's checks and permission to act.
Illustrative workflows · Not recorded judgments
Once a question makes sense in the playground, keep its definition. Export the configuration or copy the corresponding API example, then use a JevTypeSafe key to connect your own environment.
Start with an example, choose Jev or Laya, then edit the text, question, and answer definitions for your task.
Reuse the question and output definition instead of rewriting them for every entry point.
Sign in to create a key for calls from your code or Agent.
Follow the API, CLI, MCP, or skill instructions for the environment you use.
{
"model": "jev",
"questions": {
"judgment": {
"instructions": "Which team should handle this ticket?",
"type": "choice",
"criteria": {
"Billing": "Payment issues, invoices, payouts, bank account errors",
"Technical": "Bug reports, API errors, crash logs",
"Sales": "Plan upgrades, contract renewals, bulk deals",
"Account": "Password resets, team seat invites, profile updates"
}
}
}
}Use the API from an application.
Use your JevTypeSafe key with the API. See the docs for local CLI, MCP and skill setup.
The CLI, MCP, and skill clients are open source. They share the same judgment configuration and account allowance, so changing how you call Jev does not create a separate quota. Reusing a configuration in code does not bypass the output guarantees of the service.
Use your Jev playground configuration with the API, CLI, MCP, and skill →
Common questions about JevTypeSafe, Jev & Laya models, credits, and code integration.
JevTypeSafe lets you try Jev and Laya System One models in your browser. Ask for a category, a yes/no decision, or a score, inspect the calibrated response, and reuse the configuration in your own development workflow. Start with the preset scenarios if you want to see realistic inputs and criteria together.
Jev is TypeSafe's flagship System One model supporting rich documents and questions up to 8,000 tokens. Laya English (512 tokens) and Laya Multilingual (1,024 tokens) are ultra-fast, lightweight models specialized for high-throughput short text, comment moderation, and real-time LLM guardrails with sub-0.35s latency.
No. Jev models make structured judgments; they do not generate free-form text or reasoning explanations. Use Choice for named alternatives, Yes/No for a binary question, and Score for a defined scale. Your application code decides how to act on the result.
No. A probability expresses the model’s judgment for the supplied input and question. It is not a correctness guarantee. Inspect ambiguous cases, keep your own permission checks, and decide when human review is needed before taking irreversible actions.
Sign up for 10,000 one-time credits. One credit covers 1,000 input tokens across Jev and Laya models, including questions and criteria. Usage is charged exactly, including fractions of a credit. When your credits run out, buy a token pack or choose a monthly plan on the Pricing page.
You can try preset scenarios directly in the browser playground. Create a JevTypeSafe key on your account to call from your own code, CLI or Agent. Keep that key private.
Yes. Export the judgment configuration or copy its API example, then follow the developer instructions. The same question and output definition carry into your workflow across API, CLI, and MCP.
JevTypeSafe does not persist your input or result history without explicit execution. We record minimal usage and billing metadata. Your draft stays in this browser for the current tab session.
Try Jev & Laya on jevtypesafe.net with your own text, inspect the result, and keep the configuration for your workflow.
Run Jev10,000 credits when you sign up. No credit card required.