i have a room that used for recording, no ai agent joins the room, it is a human and recording the full screen, does that will consume resourses from the agent server or not ?

Thanks for reaching out, Mostafa. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

how can i debug that or which things to check to know why it consume resorces from the agent

Thanks — I’ll help you debug why the agent server appears to be consuming resources. A few quick questions first so I can narrow the steps: • Is the agent server deployed in your environment (self-hosted) or using LiveKit Cloud agents? • Do you have Agent Observability enabled for your project or agents? • Are you seeing resource consumption constantly or only at certain times (e.g., when recordings start)? • Do you have access to agent server logs and metrics (CPU, memory, process list)? If you want general checks immediately, do the following: 1. Check agent server registration and processes • Confirm the agent server is registered/connected. If it’s running, it may hold idle worker processes. • Check running processes to see idle job processes (they may be prewarmed). 1. Inspect prewarm / setup logic • Look for any

prewarm

or

setup_fnc

that loads models (VAD, ASR) at startup — these keep memory in use. 1. Look for active or pending jobs • Check agent job queue / server metrics for active or pending jobs that could be consuming CPU. 1. Review observability/recording settings • If Agent Observability or session recording is enabled, that can create extra processing and storage activity per session. 1. Verify draining / leftover deployments • Confirm no old deployments are still draining (graceful drain can keep processes alive for up to the drain timeout). 1. Review permissions / subscriptions • Ensure the agent server is not subscribing to or publishing tracks unnecessarily (restrict with

can_subscribe=False

,

can_publish=False

if applicable). 1. Enable detailed logs and metrics • Increase log level to INFO/DEBUG and check for unexpected job assignments or connections. • Monitor CPU/memory over time and correlate spikes with logs. If you share: • whether the server is self-hosted or Cloud, • recent agent logs (errors, job assignments, startup messages), and • a short timeline of when you see resource use — I’ll suggest targeted commands and exact log lines to inspect. Sources: Server lifecycle | LiveKit docs | examples/survey | livekit/livekit_composite | DeepWiki | Agent server options | LiveKit docs | VAD Metrics | LiveKit docs | Streamline Troubleshooting with Agent Observability