Voice calls, capture, and Hey Wombat

Ponder can capture microphone audio (you) and optionally system audio (another speaker on a call). Speech becomes transcript bubbles in the room chat; the agent can turn that stream into graph updates.

Microphone capture and participant voice calls are separate. Turning on the mic transcribes your speech without sharing live audio. Voice call is on by default, so everyone in the chat joins and can speak with each other; the chat owner can turn it off from the chat Settings menu. Mic on is unmute. You can remain in the call and listen with your mic off, and moving between canvas rooms does not leave the chat-bound call. Only microphone audio can be shared with participants; system audio continues to feed Ponder's transcription privately.

Voice control uses hot phrases (defaults include the product name):

Phrase	Intent
Hey Wombat	Wake the agent
over / trailing go Wombat	Submit the current voice bubble
Go Wombat	Trigger the agent
Stop Wombat	Cancel a running turn
Never mind Wombat	Cancel a wake without submitting

Typed chat always triggers the agent on send. Voice is a way to build the shared mental model while thinking out loud — not the whole product definition.