Hi team, I am running a self-hosted LiveKit instance with the Python Agents SDK and I have hit a hard wall with proactive greetings.
Environment:
-
livekit-agents==1.5.4 -
livekit-plugins-google==1.5.4 -
Model:
gemini-3.1-flash-live-preview
The Situation: I have been trying to solve this issue regularly since v1.5.1. I have been following the community discussions closely and was under the impression that the latest v1.5.4 update finally brought the necessary protocol fixes for Gemini 3.1. However, despite the upgrade and hours of debugging, the agent still refuses to initiate conversation.
The Error: When I call await session.generate_reply(), the plugin explicitly blocks it with these logs:
-
[WARNING] livekit.plugins.google: generate_reply is not compatible with 'gemini-3.1-flash-live-preview' and will be ignored. -
[ERROR] livekit.agents: failed to generate a reply: generate_reply is not compatible...
The Breaking Point: I am honestly confused and have completely lost the overview of how to solve this. I have been committed to LiveKit for my production deployment, but a basic “Speak First” greeting is a non-negotiable requirement.
If generate_reply is intentionally disabled for this model, is there a native alternative (like a direct push_text to the session or a specific setting) that works right now? I would really prefer to stay within the LiveKit ecosystem, but if I can’t get this fundamental behavior working with the newest Gemini models, I’m feeling forced to consider moving our entire stack to another framework like Pipecat.
Any clear guidance or a functional workaround would be massively appreciated.