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

# Playground

> Test Remark agent responses, intents, actions, and channel behavior before enabling changes for live shoppers.

The Playground is a testing space for Remark agents. Use it to run realistic shopper conversations, inspect what the agent used to answer, and leave feedback before enabling changes for live conversations.

Playground conversations run the same core AI behavior used by live conversations, including intent matching, knowledge lookup, product recommendations, citations, and configured actions. Intents only invoke in Playground conversations when they are **Enabled in playground**.

## Start a playground chat

1. Open **Playground** in the Remark dashboard.
2. Choose a persona if you are testing Remark chat.
3. Choose a shopper context if the shopper picker is available for your workspace.
4. Choose a channel if multiple testable channels are available.
5. Write the first shopper message.
6. Attach images or PDFs if the scenario needs them.
7. Select **Send**.

The Playground opens the new conversation after the first message is sent. Previous Playground conversations stay available in the sidebar, grouped by date.

## Personas

For Remark chat, the persona selector controls which persona the agent uses in the test conversation.

| Option         | Use it for                                 |
| -------------- | ------------------------------------------ |
| Random Persona | Let Remark choose from available personas. |
| AI personas    | Test one of your workspace's AI personas.  |
| Demo Personas  | Test with a built-in demo persona.         |

## Shopper context

When shopper context is available, you can simulate a known shopper by searching for the shopper or entering a shopper ID.

Use shopper context when the scenario depends on shopper-specific information, such as past orders, known email addresses, account history, or customer identity. If you do not choose a shopper, the Playground starts a generic test conversation.

## Channels

Playground always supports Remark chat. Additional channels appear when a connected integration is set up for inbox automation and can be tested in the Playground.

| Channel type               | What changes                                                                                                         |
| -------------------------- | -------------------------------------------------------------------------------------------------------------------- |
| Remark chat                | Tests the normal chat experience with a selected persona.                                                            |
| Helpdesk or ticket channel | Tests the agent in that external channel, including the required subject line and channel-specific message behavior. |

For ticket-based channels, the first message requires a **Subject**. The conversation header can include an **Open in...** button that links to the test ticket in the external platform.

<Note>
  Helpdesk-channel Playground sessions can create test tickets or conversations in the connected platform. Use test-safe messages and shopper context when testing those channels.
</Note>

## Attachments

Playground messages support image and PDF attachments. You can:

* Select files from the attachment button
* Drag files into the composer
* Paste supported images into the composer

The composer supports up to 10 attachments per message.

## Read a response

Each agent response can include several diagnostics that explain what happened.

| Item                    | What it shows                                                                                                      |
| ----------------------- | ------------------------------------------------------------------------------------------------------------------ |
| Detected intents        | The intent or intents the agent matched for that response, including the matching explanation.                     |
| Citations               | Knowledge documents used to support the answer. Inline citation markers appear inside the response when available. |
| Steps                   | Tools or actions the agent used while preparing the response.                                                      |
| Actions                 | Configured actions that ran, including success or failure details.                                                 |
| Thought process         | Coach output when available.                                                                                       |
| Product recommendations | Products or variants the agent recommended.                                                                        |
| Presented showcase      | A generated showcase card shared by the agent.                                                                     |
| Internal note           | A private note created for a helpdesk-style channel.                                                               |

Use detected intents first when debugging intent behavior. If the wrong intent matched, update the intent's matching criteria. If the right intent matched but the agent handled the case incorrectly, update the intent instructions.

## Regenerate responses

For Remark chat conversations, the latest agent response can be regenerated. Regeneration creates another version of that response, and the version controls let you compare previous and current outputs.

Regeneration is not available for ticket-based channels because those channels can create external messages or comments that cannot be undone. Continue the test by sending another message instead.

## Leave feedback

Use thumbs-up or thumbs-down feedback on agent responses to record what worked or what needs to change. Feedback can include tags and an optional note.

Feedback is most useful when it names the gap clearly. For example:

```txt theme={null}
The right return intent matched, but the response did not ask for the order number before offering next steps.
```

Use feedback alongside detected intents, citations, and actions to decide whether the fix belongs in [Intents](/dashboard/intents), [Persona Studio](/dashboard/persona-studio), knowledge, or an integration setup.

<Note>
  Need help interpreting Playground results? Contact `support@remark.ai`.
</Note>