Thanks — you're asking how many concurrent sessions you can run with a self-hosted agent while using LiveKit Cloud. Short answer: there’s no fixed global limit — concurrency depends on your server resources and configuration. Key facts and recommendations: • Self-hosted agent servers are not subject to Cloud agent concurrency caps. You can run multiple concurrent jobs; each job runs in an isolated process. • Practical limits depend on CPU, memory, and network. A useful estimate is ~100 MB RAM per conversation. • Recommended sizing: deploy about 1 worker per 5–10 concurrent conversations to maintain performance. • You can scale horizontally by running multiple agent worker instances; LiveKit will distribute jobs across them. • You can control when an agent stops accepting new jobs by implementing a

load_fnc

and a

load_threshold

on your self-hosted server (e.g., cap at 9 concurrent jobs). If you want, tell me: • your server CPU, RAM, and network bandwidth, and • the expected peak concurrent conversations — and I’ll suggest a concrete worker/instance count and an estimated concurrency target. Sources: Agent server options | LiveKit docs | Server lifecycle | LiveKit docs | examples/drive-thru | livekit/livekit_composite | DeepWiki