The three modes
Listen is listen. The eavesdrop controls a phone keypad would otherwise offer
for talking to one side or the other are switched off, because leaving them
live would make “listen only” a suggestion rather than a fact.
The panel’s button uses listen. Whisper and barge exist in the API
today and have no control on the screen yet.
Both parties hear a tone
A tone plays on the monitored call — the customer and the agent both hear it — as you join. Not on your leg: playing it to you would announce the monitoring to the only person who already knows about it, and the customer is the party with the least say in any of this. It plays once somebody has actually answered your handset, so the tone marks the moment you could first hear something rather than the moment you pressed the button.It leaves a mark
Your email and the mode are written into the call’s own journey, as a step anybody who later reads that call will see. A supervisor should be as visible afterwards as the call they listened to. Your leg is marked as a listener rather than a call, so it is not billed, not reported and not counted as a conversation you had.Where the button is
On the Status view beside live calls: on Queues, a queue’s own page, IVR flows and Processes. It is not on call history, which is a record of calls that have ended, and not on the desk. Stop listening hangs up your leg. Monitoring you cannot stop is not a feature.When it refuses
A call the system cannot attribute to a team is refused too, the same way it
is refused in your call history.
Playing a recording afterwards
Recordings are the other half of oversight, and the permission is separate from reading a call record: a call record is metadata, a recording is the customer’s conversation. You may play the recording of a call belonging to one of your teams’ agents — and of a conference session — and nothing else. Anything outside that reads as not found rather than as refused. Two things are commonly confused here:- A recording exists and there is audio to play are different answers. A call recorded on a node that could not write the file says so instead of offering a broken player.
- Reading a transcript of an AI call is
conversations:read, which every agent holds. Playing the audio is yours. The text answers what was agreed; the audio is a recording of a person.