https://livekit.io logo
Join Slack
Powered by
my python livekit agent worker returns the followi...
# ask-ai
r
my python livekit agent worker returns the following message when attempting to join the room on my self-hosted livekit vm:
Copy code
python -m agents.main -a mem0stt dev
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
/home/user/repos/backend/.venv/lib/python3.13/site-packages/livekit/agents/cli/cli.py:1503: DeprecationWarning: There is no current event loop
  loop = asyncio.get_event_loop()
    17:48:33.958 DEBUG  asyncio          Using selector: EpollSelector  
    17:48:33.962 DEV    livekit.agents   Watching /home/user/repos/backend/agents  
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
    17:48:39.355 DEBUG  asyncio          Using selector: EpollSelector  
/home/user/repos/backend/.venv/lib/python3.13/site-packages/livekit/agents/utils/http_server.py:13: DeprecationWarning: loop argument is deprecated
  self._app = web.Application(loop=self._loop)
    17:48:39.361 INFO   livekit.agents   starting worker {"version": "1.3.5", "rtc-version": "1.0.19"}
    17:48:39.362 INFO   livekit.agents   preloading plugins {"packages": ["livekit.plugins.langchain", "livekit.plugins.openai", "livekit.plugins.silero", "livekit.plugins.turn_detector", "av"]}
    17:48:39.366 INFO   livekit.agents   starting inference executor  
    17:48:40.985 INFO   livekit.agents   initializing process {"pid": 112805, "inference": true}
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
    17:48:47.912 INFO   livekit.agents   process initialized {"pid": 112805, "inference": true, "elapsed_time": 6.92}
/home/user/repos/backend/.venv/lib/python3.13/site-packages/livekit/agents/utils/http_server.py:26: DeprecationWarning: Application.make_handler(...) is deprecated, use AppRunner API instead
  handler = self._app.make_handler()
    17:48:48.756 INFO   livekit.agents   registered worker {"agent_name": "", "id": "AW_dhHficBPa5KE", "url": "<wss://livekit.myselfhostedvm.io>", "region": "", "protocol": 16}
    17:49:17.864 INFO   livekit.agents   received job request {"job_id": "AJ_a7fmJ3ySGuvb", "dispatch_id": "AD_qvESNXerPFdT", "room": "room-daef50db-bb9d-4f5e-973f-ff4dcba5ce75", "room_id": "RM_U6yfTTRse5mT", "agent_name": "",           
                                                              "resuming": false, "enable_recording": false}                                                                                                                                  
    17:49:18.047 DEBUG  livekit.agents   received assignment {"agent_name": "", "room_id": "RM_U6yfTTRse5mT", "room": "room-daef50db-bb9d-4f5e-973f-ff4dcba5ce75", "job_id": "AJ_a7fmJ3ySGuvb", "dispatch_id": "AD_qvESNXerPFdT",            
                                                             "enable_recording": false}                                                                                                                                                      
    17:49:18.058 INFO   livekit.agents   initializing process {"pid": 113183}
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
    17:49:22.128 INFO   livekit.agents   process initialized {"pid": 113183, "elapsed_time": 4.07}
    17:49:22.125 DEBUG  asyncio          Using selector: EpollSelector {"pid": 113183}
2026-02-20 17:49:22.148 | DEBUG    | agents.mem0stt:entrypoint:97 - [*] Starting the entrypoint
    17:49:32.152 WARNI… livekit.agents   The room connection was not established within 10 seconds after calling job_entry. This might mean that job_ctx.connect() {"pid": 113183, "job_id": "AJ_a7fmJ3ySGuvb", "room_id": "RM_U6yfTTRse5mT"}
                                         was never invoked, or that no AgentSession with an active RoomIO has been started.                                                                                                                  
    17:49:37.816 WARNI… livekit          livekit::rtc_engine:401:livekit::rtc_engine - failed to connect: Connection("wait_pc_connection timed out"), retrying...  {"pid": 113183, "job_id": "AJ_a7fmJ3ySGuvb", "room_id": "RM_U6yfTTRse5mT"}
                                         (1/3)                                                                                                                                                                                               
    17:49:38.159 INFO   livekit.agents   process exiting {"reason": "", "pid": 113183, "job_id": "AJ_a7fmJ3ySGuvb", "room_id": "RM_U6yfTTRse5mT"}
    17:49:38.158 DEBUG  livekit.agents   shutting down job task {"reason": "", "user_initiated": false, "pid": 113183, "job_id": "AJ_a7fmJ3ySGuvb", "room_id": "RM_U6yfTTRse5mT"}
    17:49:38.161 DEBUG  livekit.agents   http_session(): closing the httpclient ctx {"pid": 113183, "job_id": "AJ_a7fmJ3ySGuvb", "room_id": "RM_U6yfTTRse5mT"}
                 DEBUG  livekit.agents   http_session(): creating a new httpclient ctx {"pid": 113183, "job_id": "AJ_a7fmJ3ySGuvb", "room_id": "RM_U6yfTTRse5mT"}
t
Thanks for reaching out, Josimar. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
Thanks — this is helpful detail. I’m responding to the log lines that show the job was received but the per-job WebRTC connection fails with a 10s timeout and a 
wait_pc_connection timed out
retry. Below is a focused checklist (what the logs indicate plus concrete things to check and try). If anything is unclear, tell me which part of your agent code or which config file you want to share and I’ll walk through it with you. Key facts from your logs • Worker registers successfully with your server: 
registered worker ... "url": "<wss://livekit.myselfhostedvm.io>"
. So job dispatch/connect for control is working. • Job is received and the agent starts processing, but: ◦ “The room connection was not established within 10 seconds after calling job_entry.” — either 
job_ctx.connect()
was never invoked in your entrypoint, or it was invoked but WebRTC negotiation failed. ◦ 
failed to connect: Connection("wait_pc_connection timed out")
— the peer connection never completed (ICE/DTLS/WebRTC negotiation). What to check and try (step-by-step) 1) Confirm your entrypoint calls connect • Ensure your agent’s entrypoint calls 
await ctx.connect(...)
(or 
job_ctx.connect()
equivalent) to start the RoomIO/peer connection. • Example pattern to include:
Copy code
await ctx.connect(auto_subscribe=AutoSubscribe.ALL_ROOM)
If you don’t call this, the job will never establish the per-job media connection and will hit the 10s timeout. 2) Verify TURN/ICE and networking on your self-hosted VM • Self-hosted setups commonly fail at this stage when ICE/STUN/TURN is not available or UDP is blocked. • Ensure TURN is enabled in your LiveKit config (for a self-hosted VM you must provide a TURN/relay for NAT/firewall situations). • Open required ports on your VM/firewall: ◦ UDP ephemeral ports (for direct ICE) and TCP/443 for TURN/TLS fallback. • If UDP is blocked and TURN is not configured, the peer connection will timeout. 3) Check TLS / WebSocket access • Confirm 
<wss://livekit.myselfhostedvm.io>
resolves and the certificate is valid (not self-signed). • A valid TLS cert is required in many environments for WebRTC to fully work. • Quick test: try a wss connection from the worker host with a tool like 
wscat
to confirm handshake (or curl/openssl s_client to confirm certificate). 4) Try a browser client to the same room • Join the same room from a browser (using your server URL). If the browser also fails to connect -> problem is server/network/TURN. • If the browser can join but agent cannot -> issue is agent-side (code, environment, or missing dependencies). 5) Review agent logs for model-loading delays • Your logs show repeated: “None of PyTorch, TensorFlow >= 2.0, or Flax have been found.” • Missing ML frameworks won’t directly prevent connect(), but if your entrypoint waits on models before calling 
ctx.connect()
that could delay/hang and trigger the timeout. If your agent depends on models, preload them in a 
prewarm
or ensure 
await ctx.connect()
runs before long loads. 6) Retry/connect settings • The error shows it retries (1/3). If you want longer time to establish the peer connection in code, ensure your job logic calls 
await ctx.connect()
promptly and doesn’t block the event loop during setup. • Avoid long synchronous/blocking initialization before calling 
connect()
. If you want, I can help further with either: • Reviewing the relevant portion of your agent entrypoint (paste the function that handles job start / calls connect). • Reviewing your LiveKit 
livekit.yaml
(TURN/port/cert settings) — paste the relevant portions (mask secrets). • Commands to test connectivity from your VM (I can give specific commands to run). Which would you like to do next — share the entrypoint code that starts the job, or share your LiveKit config / network details so I can point to the next specific change? Sources: Agents Framework | livekit/livekit_composite | DeepWiki | Job lifecycle | LiveKit Documentation | Firewall Tips for LiveKit Connectivity | LiveKit | server-sdk-js | livekit/livekit_composite | DeepWiki | Agents (Node.js) | livekit/livekit_composite | DeepWiki
r
I tried using the livekit meet demo code and it works but when i try to connect my livekit agent, it connects and disconnects the agent
t
Thanks — that’s useful context. I understand the Meet demo connects fine but your agent connects briefly then disconnects. I’ll help narrow this down and get you a fix. Quick checklist of the most likely causes (based on your previous logs and the demo working) • Agent never calls 
await ctx.connect()
promptly in the job entrypoint, or it’s delayed/blocking by model loads or heavy sync work. • The agent calls 
connect()
but WebRTC negotiation fails (ICE/DTLS/TURN) and the connection times out/disconnects. • The agent’s join token is invalid or expires immediately. • The agent process is exiting or the job task is being shut down after connect (e.g., an exception after connect). • TLS or certificate issues that affect the agent but not the browser demo (different client implementations/sensitivity). To diagnose quickly, please provide one or more of the following (pick whichever is easiest to share): • The agent entrypoint function (the code that receives the job and calls 
connect()
), especially around where 
ctx.connect()
is called. • The LiveKit server 
livekit.yaml
lines for TURN and TLS (mask secrets). • The agent logs around the moment of connect/disconnect (from just before connect through the disconnect). You’ve shared earlier logs — a contiguous snippet that includes the connect attempt, any errors/exceptions, and the disconnect lines will help. • Confirm whether the agent uses the same server URL, token generation method, and certificate trust chain as the browser demo. Immediate checks you can run now 1) Confirm connect is called early • Verify 
await ctx.connect(...)
runs before any blocking model loads or long sync setup. If model loading runs first, call 
await ctx.connect()
before heavy initialization. 2) Watch for exceptions after connect • After your agent calls 
connect()
, is there any exception or call to 
process exiting
or 
shutting down job task
? If so, paste those log lines — they indicate why the job ended. 3) Token validity • Ensure the join token used by the agent is valid and not expiring immediately. If you create a short-lived token, try a token with a longer TTL to rule out immediate expiry. 4) Compare ICE/TURN behavior vs demo • Since the Meet demo works, confirm the demo and the agent use the same signaling URL and that the agent environment can reach the TURN server (if used). From the agent host, test TCP 443 and UDP connectivity to your TURN host. 5) Confirm TLS trust • If the agent runtime rejects a self-signed cert while the browser was allowed/trusted, the agent will disconnect. Confirm the agent environment trusts the certificate used by your LiveKit host. If you’re ready to share one thing now, paste your agent entrypoint function (the job start code) with any surrounding try/except and the connect call. I’ll look for: missing/late 
ctx.connect()
, blocking calls before connect, or exceptions after connect that cause the worker to shut down. If you’d rather run commands first, tell me which you prefer and I’ll give the exact commands to test connectivity and token checks.
2 Views
Open in Slack
PreviousNext
Powered by