Skip to main content
Behaviour controls how an Agent responds when it uses your Knowledge Sources.

What you will learn

  • What each Behaviour field does
  • How to write useful instructions
  • How to create safer fallback and guardrail behavior

Open Behaviour

Open an Agent and scroll to the Behaviour card.
Behaviour settings placeholder

System Prompt

Use the System Prompt to describe the Agent’s role and goal. Keep it specific. Good prompts usually include:
  • The business or team the Agent represents
  • The topics it should help with
  • The kind of answer it should give
  • When it should ask a follow-up question
  • When it should direct the visitor to support

Tone

Choose the tone that matches the audience:
  • Friendly for approachable customer support
  • Professional for sales, business, or account workflows
  • Neutral for factual help center experiences

Verbosity

Choose the response length:
  • Short for quick FAQ answers
  • Balanced for most support experiences
  • Detailed for setup, troubleshooting, or technical explanations

Fallback Message

The Fallback Message is shown when the Agent cannot answer confidently. Use a helpful message that sets expectations and offers a next step. Example:
I do not have enough information to answer that yet. Please contact our support team and include what you were trying to do.

Guardrails

Use Guardrails to define boundaries. For example, tell the Agent not to make pricing promises, not to answer outside your product scope, or to ask for a human handoff when sensitive information is involved.

Save and test

Click Save, then test the Agent in the Web Preview with real customer-style questions.
Agent preview placeholder

Next steps

Web Channel

Configure the visible web chat experience.

Best practices

Learn how to refine instructions and source content.