Playground

Chat with your agent in a test space, adjust its model and instructions, and reset conversations as you iterate.

The Playground is where you test your agent before customers see it. It works like a real conversation, with a settings panel next to it for quick changes.

Open your agent and select Playground.

Test conversations

Type a message to chat with your agent. Try:

• Common questions your customers ask.
• Questions your knowledge doesn’t cover, to see how the agent handles them.
• Off-topic or tricky questions, to check the agent stays in its role.

Select Reset Chat to clear the conversation and start fresh.

Change settings while testing

The settings panel has:

• Model: the AI model.
• System prompt: the agent’s instructions. Select Use a template… to start from an example.

Select Save to apply your changes. They’re saved to the agent, so they affect every channel, not just the Playground.

You can hide the settings panel to give the chat more space.

Things to know

• Playground conversations use credits, like live ones.
• Your agent uses the same knowledge as on your live channels.
• Actions run in the Playground as they would on your web widget, and they run for real: a Custom API action calls your API, and a Slack Notification posts to Slack. Actions turned off for the web widget are off here too. Widgets with a matching When to show condition can also appear in the Playground.
• Under each reply, a line shows the sources and action steps behind it: the knowledge it read, what ran, what it sent and received, and why other actions didn’t run.
• Escalate to Human never runs in the Playground, so test conversations don’t reach your Inbox or Helpdesk.
• There’s no website page behind the Playground, so {{page.url}} and the UTM variables are empty.
• Playground conversations appear in Logs, so you can review them later.