Posting for help. start_room_composite_egress returns not_found (“requested room does not exist”) on a room the agent is connected to and actively producing/consuming audio in. The agent runs the call end-to-end normally (audio flows, transcripts produced, TTS played out, voicemail detection fires, call ends cleanly) — only the egress recording fails.
Egress request
api.RoomCompositeEgressRequest(
room_name=<the room my agent is connected to via job dispatch>,
audio_only=True,
audio_mixing=api.AudioMixing.DUAL_CHANNEL_AGENT,
file_outputs=[...],
webhooks=[...],
)
Client SDKs: livekit==1.1.2, livekit-agents==1.4.6 (Python).
The room_name passed to egress is the same name LiveKit gave us in the SDK’s received job request event — same string, no whitespace or encoding surprises.
Retry timeline (UTC, single call)
The call ran in room ID RM_QQeiEgbwzhkb throughout. The agent was inside that room and audio was flowing through it during every one of these retries.
| Timestamp | Event | Detail |
|---|---|---|
12:05:53.159 |
livekit.agents “received job request” |
room_id=RM_QQeiEgbwzhkb, dispatch_id=AD_7ApXzCanvBAu, job_id=AJ_A4Z2GYK8rHKE |
12:05:53.694 |
First start_room_composite_egress request begins |
— |
12:05:54.157 |
Retry #1: TwirpError(code=unknown, message="context deadline exceeded", status=500) |
request took ~463 ms server-side, then timed out |
12:05:58.361 |
Retry #2: TwirpError(code=not_found, message="requested room does not exist", status=404) |
from here on, every attempt returns the same 404 |
The transition between retry #1 (status=500 "context deadline exceeded") and retries #2-5 (status=404 "requested room does not exist") looks meaningful — like the first attempt hit some server-side timeout, after which subsequent name-based lookups settled into “not found”.
Additionally — two rooms with the same name in the dashboard
While digging through the LiveKit dashboard to confirm RM_QQeiEgbwzhkb was healthy, I noticed there are actually two distinct room IDs sharing the same room name:
| Room ID | Status | Duration | Participants | Agent connection |
|---|---|---|---|---|
RM_8EGjd4UNE5Xw |
Active, “In progress” | 43 min and counting | 0 | Agent never connected to this; not logged anywhere on the agent side |
RM_QQeiEgbwzhkb |
Active | 43min and counting, actual duration ~30secs | Agent + 1 SIP participant | The room our agent actually ran the call in |
RM_8EGjd4UNE5Xw was created roughly 5 seconds before RM_QQeiEgbwzhkb, has zero participants throughout, and has been sitting “In progress” for 43 minutes with no participant ever joining. Both the rooms are being shown as active for the last 43 minutes. I know that LiveKit dashboard is
I don’t know whether this is related to the egress 404, but if start_room_composite_egress resolves by room_name and there are two same-named rooms in some active/empty state, that ambiguity seems like a plausible explanation for the first attempt’s context deadline exceeded and the subsequent not_found responses.
Please help check what exactly tool place here.
, but today we had an