Support the capability to capture a portion of the screen and then run a prompt against the captured image for models that support vision.