Skip to main content
Your AI Agent can recognize when a user sends an image, including the file name or attachment title. However, the Agent cannot see what is in the image or interpret its content. To ensure the AI Agent responds appropriately to images, you can add specific instructions in the Guardrails on how the Agent should handle them.

​
Adding an instruction in the domain knowledge

You can decide how the AI Agent should respond when someone sends an image. To do this, include a clear instruction in the Guardrails. Example:
If someone sends you an image, let them know that, as an AI Agent, you cannot view or interpret it, and therefore cannot respond to it in detail.

​
Tips for writing the instruction

  • Use clear and simple language
  • Explicitly mention that the AI Agent cannot see image content
  • Encourage the user to describe the image
Include your Handover trigger in the instruction so end users will still be helped in the chat. Example:
If someone sends you an image, let them know that, as an AI Agent, you cannot view or interpret it, and therefore cannot respond to it in detail. Ask to send “Speak to an Employee” so a human Agent can help you.