The decision playground

Jev Playground: text in, decisions to inspect.

Use the Jevai Playground on jevai.pro to test one text context with your own choices, or classify a small text file in batch mode. Inspect the returned labels and probabilities, then download batch results as CSV or JSON.

Decision workspaceLive API
Try an example

Choose an action from interface text, rather than generate a browser command. Try replacing the entire goal with a request to update a saved card.

Source: X · @milindlabs. Inspired by the author's UI-element selection demo. This is our text-only scenario; it does not control a browser or reproduce the reported timing.

Decision rules and labels
View receiptsPayment methodsManage planContact salesNo suitable control
A real response from the Jev API

Edit any input to explore a different decision. No request is sent until you run it.

Your next decision

Not run yet

Choose an example or write your own input, then run it to see the ranked choices.

Results are not stored by this site. Your submitted text is processed by the API provider.

Input contract

Define the text, question and allowed labels.

Jev is easier to evaluate when the decision boundary is visible. Use one context, one question, and a finite choice list.

  • Context: what the user or system knows.
  • Question: the decision you want the model to make.
  • Choices: the only valid outputs your workflow accepts.
Illustration of context, one question, and a finite choice list feeding a decision model that returns a choice.
One context. A finite set of choices. A result your code can inspect.
First run

Run your first decision in four steps.

Describe the situation, ask a question, define the valid routes, then inspect the result.

  1. Illustration of a delayed parcel and a customer message asking where the package is.

    Start with context

    Describe what happened in the same words your application receives.

  2. Illustration of one focused support-routing question: Which team?

    Ask one question

    State the decision clearly, so the model knows what to resolve.

  3. Illustration of four allowed team choices: Billing, Refund, Shipping, and Technical.

    Define the choices

    List the valid routes that your application can act on.

  4. Illustrative Shipping choice with ranked result bars and a handoff to application code; not actual API output.

    Use the decision

    Inspect the leading choice and score before taking the next action.

Fit check

Use Jev where choices are finite.

Good fit

Finite decisions

  • Routing requests to a known team.
  • Choosing one tool from a controlled list.
  • Flagging content for human review.
Look elsewhere

Open-ended work

  • Long-form writing with no fixed output.
  • Tasks that require open-ended generation or a multi-turn conversation.
  • Problems with no stable choice set.
Playground FAQ

Single and batch Playground questions.

Get to know the choices, confidence scores, and data handling before your first run.

What can I test in the Jev Playground?

The Jevai Playground offers single and batch text workflows. In single mode, supply context, one question and your allowed choices; the result includes a selected label, probabilities and confidence. Batch mode classifies text rows under enabled dimensions and shows row results and counts. CSV and JSON downloads are available for batch results.

Which files and batch sizes are supported?

Use CSV, TSV or TXT, or paste text. The browser file limit is 128 KiB; a batch supports up to 20 rows and 3 enabled dimensions. XLSX and image inputs are not supported here. Smaller per-row and total-text limits also apply and are checked before submission.

How should I handle uncertainty or a failed row?

Review the distribution and the row status before using a result. A high confidence value can still accompany an incorrect decision. Keep an explicit review policy, inspect failed or timed-out rows separately, and do not count them as successful classifications.

When is my input sent to the model?

Editing or loading an example does not run a decision. Submitted text is sent through this site's protected service to TypeSafe after security verification. Avoid secrets or personal data. Downloaded CSV or JSON files are assembled in your browser, while submitted requests are processed by the site proxy and TypeSafe API under their operational policies. The displayed request time includes network and service work; the cost display is an API estimate, not a bill or benchmark.

Have a real routing problem?

Start with a support example, then bring your own input here.