> ## Documentation Index
> Fetch the complete documentation index at: https://agent-docs.magicwebs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Test in the playground

> Chat with any agent as a customer would, on the same engine that serves WhatsApp, and see which tools and agents each reply used.

The playground is a private chat with one of your agents. It runs on the same engine that serves WhatsApp and the website chat, so knowledge lookups, the catalogue, record books, tools and routing all behave as they will for customers.

Open it from **Test** on an agent card, **Test in chat** in the agent editor, or **Test routing** on the front desk card.

![The playground showing Priya's opening line and suggested first messages](https://mintcdn.com/magicwebs/lXS3DuVdQz1WpiFT/images/console/agent-playground-chat.png?fit=max\&auto=format\&n=lXS3DuVdQz1WpiFT\&q=85\&s=9ac154ebf6500dad5a21bd7f260de423)

## Have a test conversation

1. Type in **Type as a customer would** and press Enter (Shift+Enter adds a new line). Or tap a starter such as *What do you sell?* or *I'd like to speak to someone*.
2. Keep going until the agent files something or hands over. In the demo, asking the front desk for "a 3 BHK near SG Highway" hands you to Priya, who offers a site visit and files it.
3. Press **Reset** to start a fresh conversation.

## Controls

| Control | What it does |
| - | - |
| **Speak replies** | Reads each reply out loud |
| **Tool trace** | Shows the tools each reply called, such as `search_products` or `file_site_visit`, and which agents it passed through |
| **Reset** | Clears the conversation |
| **Configure** | Opens the agent editor |

> [!TIP]
>
> Test from the **front desk** to check routing end to end. Test a **specialist** directly to fine-tune its replies.

## Good things to test

* Questions your FAQs answer, and a few they don't. Unanswered questions should appear under [knowledge gaps](https://agent-docs.magicwebs.ai/build/knowledge-base.md#knowledge-gaps).
* A price question for an entry with a price, and for one without. The agent should not quote the second.
* A full booking or order, to check the agent asks every required question before filing.
* An angry or urgent message, to check the agent escalates to a person.

> [!NOTE]
>
> Playground conversations appear in the [inbox](https://agent-docs.magicwebs.ai/inbox/conversations.md) as web conversations, and records filed during a test are real records. Delete or move test records when you are done. A paused agent cannot be tested; set it to Draft or Active first.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.