Skip to main content

Playground

Use Playground when you want the fastest experimentation loop inside a project.

Before You Begin​

  • You have a project.
  • You have the datasets, agents, or eval setup needed for the experiment.

When to Use Playground​

  • validate a setup quickly before saving it as a repeatable eval
  • inspect recent exploratory run history
  • generate adversarial or derivative datasets
  • check whether an agent/specification combination behaves the way you expect

Run an Exploratory Experiment​

  1. Open Playground inside a project.
  2. Review the run configuration shown on the page.
  3. Start a playground run.
  4. Wait for the run detail page to open.
  5. Review the outputs, status, and any errors on the run detail page.
  6. Return to playground to review run history and filter runs by status.

Generate Follow-On Datasets​

If the experiment produces useful transformed cases, use playground to generate an adversarial or derivative dataset so you can preserve and reuse those cases later.

This is useful when you want to turn exploratory work into a reusable input set for specifications or later comparison.

Important Notes​

  • Playground is optimized for exploration.
  • Playground run history helps you revisit recent experiments.
  • Derivative dataset generation can start from existing data and chosen attack methods.
  • Use Evals when the setup should become a saved team workflow.