Skip to main content

​
Supervisor Mode

Supervisor Mode is currently in development. Request early access from the Supervisor page in Callem Studio.

​
Overview

Supervisor Mode brings real-time human oversight to AI voice agents. Like Waymo keeps remote operators ready to assist autonomous vehicles, Supervisor Mode keeps your team ready to intervene on any live call.

​
Planned Features

​
Live Listening

Monitor any active call in real time. See the full transcript streaming live with criticality scoring.

​
Whisper Mode

Send instructions to the AI agent during a live call. The caller never hears the whisper. The agent adapts its next response based on the supervisor’s guidance. Example: A supervisor sees the agent struggling with a retention call and whispers: “Propose a 3-month discount to retain the customer.”

​
Takeover

Take direct control of the call. The AI pauses, the supervisor speaks directly to the caller, and can hand back to the AI when ready.

​
AI Criticality Scoring

An AI monitoring system continuously scores each active call on a 0-100 scale:
ScoreLevelAction
0-40StableLogged only
40-70WarningNotification in dashboard
70-100CriticalPush alert to supervisor

​
Instant Transfer

Route the call to a human agent queue with one click. Full context and transcript are passed to the receiving agent.

​
Intervention Analytics

Every whisper, takeover, and transfer is logged and becomes a training signal. Your agents improve from every human intervention, creating a continuous improvement flywheel.

​
Architecture

Call in progress
  |
  v
Voice Agent + AI Monitoring (parallel)
  |
  v
Criticality Score (0-100)
  |
  +--> Green (0-40): Log only
  +--> Orange (40-70): Dashboard alert
  +--> Red (70+): Push notification
         |
         v
    Supervisor decides:
      - Listen (passive)
      - Whisper (instruct AI)
      - Takeover (speak directly)
      - Transfer (route to human)
      - End call (emergency)

​
Why Supervisor Mode Matters

  • Enterprise trust: CTOs need assurance that AI agents have human oversight before deploying at scale
  • Regulatory compliance: The EU AI Act will require human oversight for high-risk AI systems in direct interaction with users
  • Continuous improvement: Interventions become training data, creating a flywheel where the AI improves from every human correction
  • Same level of control as traditional call centers: Supervisors in call centers have always been able to listen, whisper, and take over. Supervisor Mode brings the same capabilities to AI agents.