VoiceGateway
Get startedYour callsThe layersTeam and keysHow it worksAPI reference

Agent workers

How busy each of your agent workers is.

The layers of a call with Agent workers highlighted. The Agent workers page: agents, failed jobs, dispatch-to-join time and checks to fix, then one row per agent with where it runs, its status, CPU, memory, jobs, failures and join time.
You seeIt comes from
Load and active jobs per workerThe worker's own metrics page
Agents and versions on LiveKit CloudLiveKit Cloud
CPU and memory of a self-hosted workercAdvisor, for a container you map

Add a self-hosted worker's metrics address:

voicegateway project workers add http://agent-host:8081/metrics --project production

Map a worker's container to read its CPU and memory:

voicegateway project resources add --org default --project production --kind worker --target worker-a --container /docker/<container-id>

Container CPU and memory has the cAdvisor setup. A source that has not answered shows as awaiting samples or stale, never as idle.

Was this page helpful?

Models

Speech to text, the language model and text to speech.

Media server

Rooms and participants on the LiveKit SFU.