Enable any combination of camera, mic, and screen share. Ask the model what it sees or hears, it'll tool-call into whichever sources are on.