runVoiceAgent

runVoiceAgent(): void

Defined in: packages/realtime-livekit-agent/src/agent.ts:568

Boots the agent as a standalone worker process, separate from the application server that mints browser room tokens.

Reads LiveKit connection config from the environment, so call it from a thin entry that has already loaded .env. An invalid drain timeout or derived pending-audio budget rejects before cli.runApp, so a worker can never start with an unsafe cap and discover it mid-turn.

ServerOptions.agent is set to this module’s own resolved path, because the LiveKit CLI re-spawns the agent per job and needs a file it can import. That makes the path an invariant rather than a detail: this module must remain importable from disk at the location it was loaded from, which is why it is derived from import.meta.url and not from the caller’s entry file.

Returns

void

Example

voice-worker.mjs

import { runVoiceAgent } from '@maelstrom-co/realtime-livekit-agent';

runVoiceAgent();

Then start it with the LiveKit worker CLI verb:

node --env-file=.env voice-worker.mjs start