Playground

The Playground is your testing sandbox. It lets you chat with your agent in real time, validate responses, and experiment before rolling updates into production.

What You Can Do in Playground

• Test knowledge → Ask questions to confirm your Knowledge Base is working.
• Try actions → Trigger your defined actions and see how the agent executes them.
• Evaluate tone & style → Check if the agent responds in the right voice for your product.
• Debug quickly → Catch issues with knowledge gaps or misfiring actions before users see them.

Getting Started

1. Open Playground from the left navigation.
2. Choose the agent you want to test from the dropdown (top right).
3. Type a prompt in the chat box and press Enter.
4. Observe how the agent responds — text, action calls, or rendered UI components.

Tips for Effective Testing

• Simulate real queries → Don’t just test obvious questions; type what real users might ask.
• Cover edge cases → Try vague, misspelled, or multi-step prompts.
• Test actions end-to-end → For example, “Create a new campaign starting May 1 with a budget of 5000.”
• Check knowledge accuracy → Validate facts are coming from the right Knowledge Base entries.

Comparing to Evaluations

• Playground → Ad-hoc, exploratory testing (like a live chat).
• Evaluations → Structured, repeatable test cases that track success/failure over time.

Use Playground when you’re building and iterating quickly. Use Evaluations when you need evidence and regression testing.

Best Practices

• Keep a set of “golden prompts” handy to quickly re-test after changes.
• When you find a wrong answer, click “Train this Copilot” or update Knowledge/Actions immediately.
• Use Playground early and often — it’s faster to catch errors here than in Conversations later.

✅ With Playground, you can safely experiment, debug, and fine-tune your agent in real time before exposing it to users.