I understand that seeing a high process memory usage warning can be concerning. This warning pops up because the memory used by a process has gone beyond the set threshold, which is currently at 500 MB. This often happens with resource-heavy features like Background Voice Cancellation (BVC), multilingual turn detection, or when using large language models. Here are some friendly suggestions to help you manage this: • Adjust memory thresholds: You might want to increase the

job_memory_warn_mb

and

job_memory_limit_mb

settings in the WorkerOptions to better fit your memory needs. • Upgrade infrastructure: It’s a good idea to use at least 4 CPU cores and 8GB of RAM per worker for smoother performance. Also, try to avoid free hosting tiers or AWS Fargate for production deployments to keep things stable. • Optimize model usage: Consider using smaller models or reduced quantizations (like Q4 instead of Q5) to reduce memory consumption. Just so you know, if the memory usage goes beyond the limit (which is currently set to 0, meaning no limit), the process will be automatically stopped to keep your system stable. Please feel free to reach out if you’d like any help walking through these steps—I’m here to support you! Sources: livekit-agents/livekit/agents/ipc/inference_proc_executor.py | Operational Guidance | livekit/livekit_composite | DeepWiki | Deploying to custom environments | LiveKit Docs | tools/main/README.md