What are AI agents in Whippy?

Last updated: November 5, 2026

AI agents are configurable assistants that can handle conversations across chat, voice, or both. Instructions control how the agent behaves, while knowledge bases provide facts and tools allow the agent to take structured actions.

Key parts of an AI agent

  • Agent instructions: Define tone, behavior, priorities, decision-making, and when to use knowledge or tools.

  • Channels: Determine where the agent operates, such as messaging inboxes or phone numbers.

  • Knowledge bases: Provide approved factual information for the agent to reference.

  • Tools: Let the agent retrieve information or perform configured actions.

  • Agent settings: Control voice, transcription, language, responsiveness, and call behavior.

  • Versions: Preserve saved changes so you can review or roll back configuration.

How AI agents work

  1. Create an agent with a name and description.

  2. Write clear agent instructions.

  3. Optionally attach knowledge bases.

  4. Optionally configure tools.

  5. Assign the agent to chat, phone, or both.

  6. Publish or activate the agent.

  7. During conversations, the agent uses its instructions to decide whether to answer, reference knowledge, or use a tool.

How instructions, knowledge bases, and tools work together

  • Instructions are the control layer for the agent.

  • Knowledge bases provide information but do not define behavior by themselves.

  • Tools provide actions but do not run automatically unless instructions guide when to use them.

Tips

  • Write instructions before relying on knowledge bases or tools.

  • Be explicit about when the agent should reference knowledge.

  • Define exact conditions for tool usage.

  • Test agent behavior after every meaningful change.

Troubleshooting

  • The agent ignores tools: Add tool usage logic to the instructions and confirm the tool is enabled.

  • The agent does not reference knowledge: Add guidance that tells the agent when to use the knowledge base.

  • The agent gives inconsistent responses: Simplify and clarify the instruction hierarchy.

  • The agent behaves differently across channels: Add channel-specific guidance for chat and calls.