Thanks — I’m responding to your question about the

AgentServer

class in

livekit.agents

. What it is •

AgentServer

is the server-side component that registers an agent with LiveKit, waits for job/room requests, and orchestrates agent sessions. It manages the lifecycle of agent jobs and acts as the entrypoint for running agent logic when a room or job is dispatched to the agent. Key responsibilities and features • Agent registration & lifecycle: registers the agent with LiveKit and spawns a job process when a request is accepted. • Entrypoint decorator: use

@server.rtc_session()

in Python to define the function that runs for each accepted job/session. • Per-job isolation: each accepted job runs in its own process so multiple sessions can run concurrently without interfering. • Configuration options: ◦

permissions

to control publish/subscribe/data capabilities. ◦

drain_timeout

to control how long the server waits for active jobs to finish on shutdown (default: 30 minutes). ◦

setup_fnc

for prewarm initialization (e.g., loading models). ◦

load_fnc

and

load_threshold

for custom load monitoring (note: not available on LiveKit Cloud). ◦

host

and

port

for the health check endpoint (defaults to

0.0.0.0:8081

). • Health checks: exposes an HTTP health endpoint at

/

for monitoring. • Dispatch control: supports automatic and explicit dispatching (routing by

agent_name

). • Graceful draining: stops accepting new jobs on shutdown and waits up to

drain_timeout

for active jobs to finish. How it’s commonly used • Pair

AgentServer

with

AgentSession

(the per-room/session component) to handle the real-time media pipeline (STT → LLM → TTS). • Start the server with

agents.cli.run_app(server)

. If you’d like, I can: • Show a minimal Python example that defines an

AgentServer

and a

@server.rtc_session()

handler. • Explain how to set specific configuration options (e.g.,

permissions

,

drain_timeout

, or

setup_fnc

). • Help troubleshoot an error you’re seeing when starting an

AgentServer

. Which of those would be most helpful right now? Sources: Server options | LiveKit Documentation | Server lifecycle | LiveKit Documentation | Build Your First AI Voice Agent in Python: Complete Tutorial | LiveKit | Deployment reliability on LiveKit Cloud | LiveKit | How to Use the Supervisor Pattern for Multi-Agent Voice AI Systems | LiveKi..., with interactive elements