We saw multiple problematic agent sessions today and would appreciate help understanding the expected LiveKit behavior here. User A: https://cloud.livekit.io/projects/p_4x0e5lq1oqj/sessions/RM_JsXtV5JsX5ZU Observed behavior: - The user joined at 2026-04-03 213742Z, and agent

agent-AJ_td4xUYphpdNU

joined at 213743Z. - The agent later left with

CONNECTION_TIMEOUT

at 215426Z. - The user left at 215432Z and then made several rejoin attempts. - We saw a brief reappearance of the same agent identity at 215521Z in the same room, but it did not recover into a usable session, and the room ended almost immediately after. - Subsequent user rejoins did not result in a healthy agent being available. User B: https://cloud.livekit.io/projects/p_4x0e5lq1oqj/sessions/RM_VsNjfGTHNMQW Observed behavior: - The agent

agent-AJ_yo2MwzgsNcYA

left with

CONNECTION_TIMEOUT

at 2026-04-03 215441Z. - After that, the room remained in a no-tutor state while the user repeatedly retried joining. - The stale room ended at 222306Z. - A fresh replacement agent only appeared after that room had ended. Our setup: - We use token-embedded

RoomAgentDispatch

so that LiveKit dispatches an agent when the user connects. - We also configure

delete_room_on_close=true

. Our questions are: 1. After an agent participant leaves with

CONNECTION_TIMEOUT

, should LiveKit automatically redispatch a replacement agent to the active room? 2. On a later user reconnect to a room that currently has no agent participant, should token- embedded

RoomAgentDispatch

trigger a fresh dispatch? 3. Does

delete_room_on_close=true

only apply when the AgentSession closes cleanly, as opposed to an agent participant timing out? 4. Can you inspect any LiveKit-side control-plane logs for these sessions to explain why no stable replacement agent was brought up?

Thanks — I’ve reviewed the two sessions and your questions about agent redispatch, reconnect behavior, and room deletion. I’ll answer each question directly and then ask one small follow-up so I can inspect control-plane logs for these specific sessions. 1) After an agent participant leaves with CONNECTION_TIMEOUT, should LiveKit automatically redispatch a replacement agent to the active room? • No. An agent participant timing out does not by itself guarantee an automatic replacement dispatch. 2) On a later user reconnect to a room that currently has no agent participant, should token-embedded RoomAgentDispatch trigger a fresh dispatch? • Not always. Token-embedded RoomAgentDispatch can trigger a dispatch when a new linking condition is met (for example, when a user joins and the RoomAgentDispatch logic decides to create/link an AgentSession). If an existing AgentSession is still considered active or in a terminal state that prevents redispatch, a fresh dispatch may not occur on the reconnect. 3) Does delete_room_on_close=true only apply when the AgentSession closes cleanly, as opposed to an agent participant timing out? • Yes.

delete_room_on_close=true

applies when the AgentSession closes. A participant timing out does not necessarily close the AgentSession; therefore the room may remain and not be deleted immediately. 4) Can you inspect any LiveKit-side control-plane logs for these sessions to explain why no stable replacement agent was brought up? • I can — to investigate the control-plane logs for those specific sessions I need one thing from you: ◦ Please confirm the full Project ID (you included p_4x0e5lq1oqj) and confirm the exact Session IDs you want examined (you listed RM_JsXtV5JsX5ZU and RM_VsNjfGTHNMQW). Also tell me if you want me to focus on a specific time window beyond the timestamps you shared (default: the join/timeout/rejoin window you listed). Once you confirm that, I’ll pull the control-plane logs for those sessions and report back what caused the AgentSession state to remain without a usable agent and why no replacement was created. , with interactive elements