Platform
Real-time guidance the customer cannot hear
A private audio channel directly to the agent's ear. The customer's WebRTC consumer is never subscribed to the whisper producer — separation is enforced at the SFU, not the UI.
<150ms
Whisper voice latency to agent
0 audio bleed
Into customer downlink
1 click
Monitor → whisper switchover
Multi-coach
Several supervisors can co-coach
Whisper uses a dedicated mediasoup audio Producer subscribed only by the agent's Consumer. The customer's downstream RTP packets never include the supervisor's audio frames — there is no client-side filter to bypass. If the supervisor is unmuted and the agent is in the room, only the agent hears them.
Switch between push-to-talk (good for high-risk regulated calls) and open-mic (good for new-agent ramp-up). Visual indicators on the agent UI show whisper status without revealing the supervisor's identity to the customer.
New agent ramp-up: a senior agent coaches a junior in real time on the first 50 calls. Compliance interventions: a QA lead pushes mandatory disclosure language during a regulated call. Multilingual escalation: a Hindi-speaking supervisor whispers translations to a Tamil-speaking agent handling a Hindi customer.
Two supervisors can whisper into the same session at the same time — for example, a product expert and a process owner during an escalated call. Each whisper producer is isolated; the agent hears both, the customer hears neither.
No. The customer's WebRTC consumer is never given a subscription to the whisper producer. Even if the supervisor's microphone is loud, the audio frames are not in the customer's downstream RTP stream.
Yes — the agent dashboard shows a small badge with the supervisor's name and a 'whisper active' indicator. The customer view is unchanged.
Yes. Whisper audio can be captured as a separate track in the session's dual recording, or excluded entirely. This is configurable per queue.