Skip to main content
The vision tool captures a snapshot from the user’s camera feed and adds that visual context to the LLM context during a live conversation. Use it when your assistant needs access to the user’s live video stream.

​
Create a vision tool

​
Attach via JSON

Add a function with type: "vision" and define a clear name and description. The function name and description are shown to the LLM as tool metadata, so they should clearly describe when the tool should be called. 
{
  "name": "<vision_tool_name>",
  "description": "<description>",
  "type": "vision"
}
For example: 
{
  "name": "extract_insurance_card_details",
  "description": "Read the current camera frame and extract insurance card details such as payer name, member ID, and group number.",
  "type": "vision"
}

​
Attach via UI

  1. Open your scenario and go to the target node.
  2. Click Add function.
  3. In the modal, select the Vision Tool tab.
  4. Enter a semantically meaningful function name and description.