generate_reply timed out waiting for generation_created event in LiveKit Agents (Gemini Live 2.5 Flash with Vertex AI) randomly.

[aankitroy]

Bug Description

Description When using Gemini Live 2.5 Flash models via Google Vertex AI with LiveKit Agents, the agent frequently throws the following error during response generation:

generate_reply timed out waiting for generation_created event This occurs after calling generate_reply() on an active LiveKit LLM session, particularly when connected to a Vertex AI gemini-live-2.5-flash model through the google-vertexai provider.

Expected Behavior

The LiveKit agent should receive the generation_created event from the Vertex AI Gemini model and begin streaming a response.

Reproduction Steps

Steps to Reproduce
Configure a LiveKit Agent to use:
Provider: google-vertexai
Model: gemini-live-2.5-flash
Livekit Agent version: 1.2.17 
Start a standard session and attempt to generate replies multiple times.
Observe that:
In some runs, the agent streams replies normally.
In others, it hangs for several seconds and then times out with:
generate_reply timed out waiting for generation_created event
The error occurs even with stable network conditions and identical setup parameters.

Operating System

Ubuntu 24

Models Used

No response

Package Versions

LiveKit Agents version: 1.2.17

Google GenAI SDK version: google-genai >= 1.41.0

Python version: 3.13

Model: gemini-live-2.5-flash

Provider: google-vertexai

Session/Room/Call IDs

No response

Proposed Solution

Additional Context

No response

Screenshots and Recordings

No response