How the call works
1
Joining
The agent joins the call and, in Announce mode, speaks the join disclosure once. If the disclosure text is left empty, it says: “This meeting is being recorded and summarized by an automated notetaker.” In Silent mode it joins without speaking.
2
Listening
The agent transcribes everything and stays muted. Recorded audio, such as phone menus, hold music and voicemail greetings, never makes it speak.
3
Ending
The agent does not hang up or say goodbye on its own. It leaves when someone says the wake word followed by a leave command, for example “Scribe, leave” or “Scribe, you can go now”. It replies “Leaving now. Summary to follow.” When the call ends, the summary is sent to the places you configured.
Talking to the Notetaker
The leave command only works with the wake word in front of it, so an ordinary sentence containing “leave” never dismisses the agent.
Pausing is immediate and private. While off the record, the agent stops sending audio for transcription and discards what it had not yet transcribed. It only listens, on its own, for the resume phrase. If that local listener cannot run, the agent says “I can’t hear voice commands right now, so notes stay off for the rest of this call.” and stays off the record.
Creating a Notetaker Agent
Choose Notetaker in the new-agent wizard, in the Dashboard or in Connect. The settings are on the agent’s settings page afterwards:
Settings meant for conversational agents are hidden or turned off for a Notetaker: the AI greeting, transfers and warm transfer, the email tool, caller detection, screenshare, filler phrases, and background and thinking sounds. Recording is always on, because the summary is built from the transcript.
The summary
When the call ends, the agent writes a summary with these sections:
The webhook receives a JSON body with
agent_id, transcript_id, the full text as summary, and each section above as its own field. Email recaps arrive with the subject “Meeting notes from” followed by the agent’s name.