How do I write effective voice agent instructions in Whippy?

Last updated: November 5, 2026

Write structured voice agent instructions so your AI agent follows the right workflow, uses tools correctly, and escalates when needed.

Before you start

  • Define the agent's role, caller types, and workflows.

  • List any tools the agent can use and when each tool is allowed.

  • Identify the knowledge sources the agent should rely on.

  • Decide when the agent should clarify, escalate, or end the call.

Steps

  1. Start with an Identity section that defines who the agent is and how it should sound.

  2. Add Objectives that describe what the agent should accomplish.

  3. Write a Primary Workflow with numbered steps.

  4. Add Intent Handling sections for common caller intents.

  5. Define Tool Usage with the tool purpose, trigger conditions, required information, and rules.

  6. Define Knowledge Base Usage so the agent knows when to use approved sources.

  7. Add Rules and Guardrails to prevent guessing, over-sharing, or unsupported actions.

  8. Add Conversation Guidelines for tone, length, and caller experience.

  9. Add Fallback Logic for unclear requests, missing information, or tool failures.

  10. Add Escalation Rules for human handoff scenarios.

  11. Add Call Ending Rules so calls close cleanly.

  12. Configure data parsing fields if you need structured reporting from calls.

Tips

  • Use Markdown headings to separate sections.

  • Break work into atomic instructions, such as one ask or action per bullet.

  • Name tools explicitly and define exactly when to use them.

  • Require confirmation before transfers, SMS sends, or other important actions.

  • Tell the agent what to do when it does not know the answer.

Troubleshooting

  • If the agent gives inconsistent responses, add clearer workflows and atomic steps.

  • If the agent makes up answers, require approved knowledge sources and escalation when unsure.

  • If tools are used incorrectly, define each tool's purpose, triggers, inputs, and rules.

  • If calls do not end cleanly, add explicit closing steps and examples.

  • If the agent gets stuck, add clarification and escalation paths.