Thanks — I’m responding to your question about allocating more memory to a LiveKit Cloud server. Short answer: increasing memory can help, but whether you should depends on the agent workload, observed memory bottlenecks, and cost tradeoffs. Here’s a practical way to think about it and what to do next. Key points to consider • Workload matters: features like enhanced noise cancellation, real‑time transcription, or running larger LLMs increase RAM needs. For typical voice AI agents, a good baseline is 4 CPU cores and 8GB RAM per agent server; that generally supports ~10–25 concurrent jobs depending on processing complexity. • Autoscaling and headroom: LiveKit Cloud autos-scales agent servers. More memory gives each server more headroom before autoscaling triggers, which can reduce latency spikes during traffic surges. • Long‑running sessions & deployments: extra memory helps agents handle long or complex conversations without running into OOMs during rolling updates or graceful drains. • Cost vs benefit: more memory increases hosting cost. LiveKit Cloud pricing bundles hosting/compute/bandwidth into session-minute charges, so avoid over‑provisioning unless monitoring shows it’s needed. • Observability first: use CPU/memory/latency metrics to confirm memory is the bottleneck before changing instance sizes. Often optimizing agent processes (model loading, batching, or multi‑tenant strategies) yields better ROI than simply increasing RAM. Recommended next steps 1. Monitor current usage: ◦ Track per‑agent memory usage, swap activity, and session latency during peak load. ◦ Identify if OOMs, garbage collection spikes, or swap thrashing are occurring. 1. If you see memory pressure: ◦ Increase RAM incrementally (e.g., from 8GB → 12–16GB) and retest under load. ◦ Consider increasing CPU alongside RAM for CPU‑heavy models. 1. If memory isn’t the clear bottleneck: ◦ Optimize agent logic (model lifecycle, batching, lower precision models). ◦ Use multi‑tenant agents or change load thresholds so scaling occurs earlier. 1. Use autoscaling settings: ◦ Tune `load_fnc`/`load_threshold` so servers spin up before memory/CPU saturate. If you want, tell me: • What your current agent size is (CPU/RAM) and typical concurrent sessions, • Which features your agents run (transcription, noise cancellation, LLMs), and • Any specific symptoms (OOMs, latency spikes, failed requests). With those details I can recommend a concrete memory/CPU target and a test plan to validate it. Sources: Self-hosted deployments | LiveKit Documentation | Deploy and scale agents on LiveKit Cloud | Can you increase agent deployment limits? | LiveKit | How much can LiveKit scale? | LiveKit