Thanks — I see you're hitting repeated "initialization timed out, killing process" errors in production. That usually means an agent worker failed to finish startup within the allowed time and was terminated. A common cause in production is CPU throttling on burstable instance types (for example t3/t4g). Loading models (VAD/LLMs) during startup is CPU‑intensive and can exceed short timeouts when the instance is throttled. To narrow this down and give a concrete fix, can you provide the following? • Your production instance type (e.g., t3.medium, t4g.small). • Whether you use a prewarm/setup function (

setup_fnc

) or

load_fnc

to load models before handling jobs. • Your agent server configuration values for timeouts (for example

drain_timeout

) or any custom initialization timeout you changed. • LiveKit SDK version and the agent language (Python or Node.js). • Any recent changes to deployment (instance sizes, autoscaling, container limits, or CPU limits/requests). If you prefer, tell me one or two of these now and I’ll suggest the most likely fixes and immediate mitigations. Sources: Agent server options | LiveKit docs | Troubleshooting Latency and Timeout Errors with Turn Detection on AWS | Still Have Questions