Any clue what this error could be caused by? It's only happening on one of our developers' devices, with identical hardware. Using docker compose as well, so it's rather strange. livekit-agent-be | {"message": "failed to handle event\nTraceback (most recent call last):\n File \"/usr/local/lib/python3.11/site-packages/livekit/plugins/openai/realtime/realtime_model.py\", line 770, in _recv_task\n self._handle_response_audio_delta(\n File \"/usr/local/lib/python3.11/site-packages/livekit/plugins/openai/realtime/realtime_model.py\", line 1356, in _handle_response_audio_delta\n item_generation = self._current_generation.messages[event.item_id]\n ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^\nKeyError: 'item_C1qpMgiAkcy28Mdb3m12m'", "level": "ERROR", "name": "livekit.plugins.openai", "event": {"type": "response.audio.delta", "event_id": "event_C1qpNsx21Apt5YZfrW2oi", "response_id": "resp_C1qpMwm2TY7YlY5UTUZDn", "item_id": "item_C1qpMgiAkcy28Mdb3m12m", "output_index": 0, "content_index": 0, "delta": "BAIDAv0B+A..."}, "pid": 27, "job_id": "AJ_VL8T54wMs2hh", "timestamp": "2025-08-07T090353.925790+00:00"}

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

Anything more helpful that you could share? Sounds like you want me to dive deep in the internals of livekit-agents & the openai plugin?

Hey there! To help you out without diving too deep into the internals, here are some practical debugging steps you can try: • Enable detailed logging: Try increasing the logging level for the LiveKit agent and OpenAI plugin. This way, you’ll get more context around what’s happening right before the error pops up. • Check event sequences: Keep an eye on the sequence of events (like

response.output_item.added

and

response.audio.delta

) to make sure they’re coming in the order you expect. • Validate session state: Double-check that the session and generation states are set up correctly and stay consistent, especially after any reconnections. • Reproduce the issue: Since this seems to happen only on one developer’s device, it might help to compare their setup (like Docker configuration or network conditions) with others to spot any differences. • Temporary workaround: You could add a quick check in the code to see if

item_id

exists in the dictionary before trying to access it, and log any missing IDs to investigate further. Hopefully, these steps help you get closer to the root cause without needing big changes to the code. If you want, I’m here to guide you through any of these—just let me know!

Where would you recommend taking these steps? Inside the livekit openai plugin?

Hey there! So, the steps should mainly zoom in on the LiveKit OpenAI plugin since that’s where the error is popping up—in the

realtime_model.py

file. Here’s what I’d suggest: • Enable detailed logging: Tweak the logging setup inside the LiveKit OpenAI plugin to get more detailed logs. This will help us see what’s really going on. • Check event sequences: Add some logging in the event handling methods like

_handle_response_output_item_added

and

_handle_response_audio_delta

so we can follow the order and details of the events. • Validate session state: Take a close look at how

self._current_generation

and its

messages

dictionary are initialized and updated inside the plugin. • Temporary workaround: You might want to update the plugin to check if

item_id

exists in the dictionary before trying to access it—and log any missing IDs so we can dig deeper later. It’s best to make these changes in a safe space like a dev or staging environment, just to keep your production system running smoothly. If you’d like, I’m happy to help you pinpoint exactly where to tweak or test in the plugin—just let me know!