Thanks — you're asking how to forcefully update an agent while its current agent process or task is still running. I’ll clarify what’s possible and ask a couple quick questions so I can give the most useful guidance. What’s currently supported • LiveKit Cloud deploys new agent versions using graceful draining: the server stops accepting new sessions and lets existing sessions complete instead of terminating them immediately. • There is no documented mechanism to force-terminate running agent sessions mid-execution during a LiveKit Cloud deployment. The system is designed to preserve active sessions and avoid disruption. • You can instantly roll back to a previous version if an update causes problems. If your goal is to make updates take effect immediately, here are practical approaches • Design agents to be stateless or to check for updates at safe breakpoints so a new version can be picked up without aborting active work. • If you control the deployment environment (for example, a custom Kubernetes cluster), you can implement a forced restart/kill of agent processes yourself — but that will terminate active sessions and may impact users. • Use rollbacks to revert problematic deployments quickly if immediate fixes are needed. A few questions so I can give targeted steps 1. Are you using LiveKit Cloud or a self-managed deployment (for example, Kubernetes on your infrastructure)? 2. Do you need to terminate running sessions immediately (downtime acceptable), or do you need a safe transition with minimal disruption? 3. If self-managed, do you want commands/manifest examples for force-restart (Kubernetes pod drain/delete) or for implementing graceful restart behavior in the agent code? Tell me which environment and which outcome you want (immediate stop vs. graceful transition), and I’ll provide concrete steps you can follow. Sources: Deploy and scale agents on LiveKit Cloud