# Playground

Chat with your agent in real time, exactly as your users would, to test responses after any change.

**Path:** Agent Studio → Playground

The Playground lets you chat directly with your agent exactly as your users would. Use it to test responses after making any change to Knowledge, Actions, or Style & Behavior — before those changes reach real users.

<PlaygroundMock />

## What you'll see

- A chat interface showing your agent's name and icon.
- Pre-filled **starter prompts** (for example, "Create a new project" or "Upload materials from invoice") — these come from your [Starters configuration](/user-guides/building-your-agent/).
- A text input at the bottom to type any question.
- A **`+`** button in the top-right to start a fresh conversation.

If you manage more than one agent, switch between them using the dropdown next to the agent name in the top bar.

## How to use it

1. Type a question a real user might ask, or click one of the starter prompts.
2. Review the agent's response for accuracy and tone.
3. If the response is wrong, the fix is usually in **[Knowledge](/user-guides/knowledge-base/)** (missing content) or **[Actions](/user-guides/authoring-actions/)** (missing capability).

Below each agent response you can give feedback (👍 / 👎), copy it, or open **View Analysis** to inspect how it was generated — the **Knowledge** tab shows which articles were referenced (with relevance scores), and the **Actions** tab shows every backend action the agent called, with its arguments, result, status, and timing.

## Testing actions with a mock response

When the conversation reaches a point where the agent must perform a task, it calls an [Action](/user-guides/authoring-actions/). In the Playground this runs as a **mock backend workflow**, so you can test safely without touching real systems.

<PlaygroundAction />

The Playground shows the action it's about to call and the properties it extracted, then pauses. Edit the **mock response** to simulate any backend outcome — success, an error, empty data — and click **Continue With This Response**. The agent processes the result and carries on, so you can validate the whole flow end to end before going live.

:::caution[What you can't test in the Playground]
The Playground runs **inside the Foldspace UI** against a mock backend, so two things can only be verified on your own site (a staging or dev build with the Foldspace SDK embedded):

- **Real-time support handoff into a chat widget** — opening the Intercom Messenger or Zendesk widget happens on *your* page, not inside the Foldspace UI, so it won't fire here. See [Support handoff](/user-guides/support-handoff/).
- **Integrations against real backend data** — actions run against the **mock response** you edit above, never your live systems.
:::

:::tip
Always test in the Playground after editing Knowledge, Actions, or Style & Behavior before pushing changes live. To lock in conversations that must always work, turn them into repeatable tests in [Trust Lab](/user-guides/trust-lab/).
:::