Playground

Test an agent before sending real customer messages.

The playground runs an agent without sending messages to real customers.

Use it to test:

• Instructions.
• Knowledge quality.
• MCP tool selection.
• Reply tone.
• Guardrails.
• Long answer formatting.

What playground runs include

Playground runs use the saved agent configuration:

• Instructions.
• Model.
• Tools.
• Integrations.
• Knowledge.
• Context profiles.

Playground runs are saved into run history so you can inspect output and timing.

What to verify

Before putting an agent on a live line, test:

1. A normal greeting.
2. A private-data request that requires MCP.
3. A request the agent cannot verify.
4. A follow-up that depends on recent conversation.
5. A long list response.

For private-data questions, make sure the run actually made an MCP call and did not ask for user-supplied identifiers first.