Skip to main content
The platform admin can monitor every active call across all tenants and users from /admin/monitoring — a platform-wide version of the tenant live monitor, with no tenant filter.

​
What it shows

  • All active calls (queued, ringing, in_progress) across the whole platform, each enriched with the assistant name, tenant name, and the owner user’s email.
  • Latency insights — platform-wide p50/p95 for end-to-end latency, LLM time-to-first-token, TTS time-to-first-byte, and STT, over the last 24 hours (from call_turn_metrics).
  • Admission control — starting and active reservations, denials over the last 24 hours, global health, and platform/region/capability controls. A missing global control is treated as unavailable.

​
Admission control

Admission control is the internal safety boundary before a new realtime session starts. It atomically protects workspace and campaign concurrency, then checks global, regional, channel, mode, and provider capacity. Pre-dispatch reservations expire automatically; terminal calls release their slots. Platform admins can change health, admission state, a hard capacity limit, and an operational reason. Changes can affect new sessions immediately, so the UI requires an explicit confirmation and records the operator, resulting state, and update time. These controls are intentionally shown only on /admin/monitoring — never on tenant monitoring, campaign, or history pages.

​
Supervisor actions

The same realtime mechanism as the tenant-level monitor is available for any active call, without tenant scope:
  • Listen in — a hidden, subscribe-only participant joins the room.
  • Barge in — join audibly and speak (visible + publishing).
  • Whisper — send a private prompt to the agent.
  • Take over — end the agent’s turn and hand the call to the supervisor.

​
Auth & audit

Access requires a platform admin (requirePlatformAdmin); impersonation is ignored and there is no plan gate — the platform admin can monitor everyone. Every action is written to the admin audit log.
  • GET /api/admin/monitoring — active calls + latency aggregates.
  • POST /api/admin/monitoring — supervisor actions (listen_token, whisper, end_agent, hangup) and the internal set_capacity_control operation.
This is a platform-admin-only tool and is not part of the public v1 API or MCP.