The other side knows we're there.
Sotto's bot is always a visible, honestly named participant in the call — never hidden, not as an option, not as a setting.
It is on screen, by name
The bot joins as an ordinary participant, labelled "Sotto — AI assistant", camera off. Every other person in the call sees its tile the same way they see anyone else's. There is no configuration that removes it from that view — the platform's underlying video provider supports doing exactly that, and this product does not use it. That is a decision, written down and standing since 2026-06-12, reconsidered twice since and held both times.
It asks before it listens
Each participant agrees before the bot subscribes to their audio track — an affirmative act on the pre-join screen, checked at the moment the bot would start listening, not left to the interface to enforce later. Nobody's speech reaches the transcript before that.
It publishes nothing back
The bot subscribes to audio; it never sends audio or video of its own. There is no code path in this product that lets it speak into the room.