The Playground is where you test prompts against your guardrails and see how they are evaluated before any real traffic flows. You send a prompt, it runs through a profile the same way live traffic would, and you see the signals detected on it. It is the place to confirm that a profile behaves the way you expect, to check that detections like PII or toxicity fire on the inputs you would expect them to, and to debug a profile's behavior on specific prompts.

The Playground tests one prompt at a time. To test many prompts at once, use Cannon, which runs a whole collection through a profile in bulk.

Getting Started

The Playground landing page offers two ways in. Send Test Prompts takes you into the prompt input so you can run your own prompts. Browse All Events shows the full Playground event history across all profiles without sending anything new. In this view the message input is replaced by a prompt to choose a profile or use case before you send anything.

Send a Test Prompt

Before you send a prompt, you select what to test it against. The filter drawer prompts you to choose a use case or profile, which determines the guardrails the prompt runs through. If you choose a use case, the prompt runs through the profile assigned to its User-to-LLM event type, or the default profile if none is set. Once you have made a selection, close the drawer to reveal the message input at the bottom of the page. The Playground sends only User-to-LLM traffic.

Type a prompt into the message input and press Enter or click the send button. The prompt runs through the selected guardrails, and the result appears in the event table above the input. You can keep sending prompts, and each one is added to the table.

Whether you get a response back depends on whether a model is connected. A profile in listen mode evaluates the prompt and shows the signals detected on the input, but does not return a response from the LLM. To get the model's response back along with the signals detected on it, the profile needs a model and must be in enforce mode.

Reviewing Results

Results populate an event table, one row per prompt you have sent. This is the same panel described in the Event Details Overview. See the Monitoring Activity Overview for how to read and configure it.

Clicking an event opens the event detail panel, which shows the full detail for a single prompt: the prompt and response, the input and output context, and the signal results. This is the same panel described in the Monitoring Events Overview, with one addition specific to the Playground.