The available sizing guidelines and benchmarks for LiveKit agent services focus primarily on higher-spec machines than yours. The official recommendation is 4 cores and 8GB RAM for every 25 concurrent sessions as a starting rule for most voice-to-voice applications. In benchmarking, a 4-core, 8GB RAM machine handled about 30 concurrent sessions, utilizing nearly all available CPU resources. There is no official benchmark or documentation detailing the specific concurrent user capacity for a 2 vCPU, 4GB RAM machine. While such a configuration may handle a proportionally smaller load than the documented baseline, you would need to run your own load tests to determine the exact limits and ensure acceptable performance at your required concurrency. Refer to your application's load profile and required latency targets, as actual performance will further depend on audio complexity, networking, and your agent logic. For higher reliability, consider following the documented 4-core/8GB per ~25 concurrent sessions rule and scaling accordingly. Sources: Deploying to production | LiveKit Docs | kubernetes/agent-manifest.yaml