Can you have multiple Workers in LiveKit? If so how can you initialize them in python with

cli.run_app()

?

Thanks for reaching out, Antonio. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

How can I have workers with different agent names?

Hey there! If you want to assign different agent names to workers in LiveKit, you can do this by setting the

agent_name

field in the

WorkerOptions

when you initialize the worker. This way, you get explicit control over which agent joins a specific room. Here’s a quick example to help you out:

if __name__ == "__main__":
    cli.run_app(
        WorkerOptions(
            entrypoint_fnc=entrypoint,
            agent_name="my-agent-1",  # Just pop in a unique name here
        ),
    )

You can spin up multiple workers with different names like

"my-agent-1"

,

"my-agent-2"

, and so on, each running as its own process or service. After you’ve got your workers named, you can use the LiveKit API to send a specific agent to a room like this:

import asyncio
from livekit import api

async def dispatch_agent():
    lkapi = api.LiveKitAPI()
    dispatch = await lkapi.agent_dispatch.create_dispatch(
        api.CreateAgentDispatchRequest(
            agent_name="my-agent-1",  # Make sure this matches the WorkerOptions name
            room="my-room",
            metadata="optional metadata"
        )
    )
    print("Agent dispatched:", dispatch)

Just a few things to keep in mind: • Each worker needs its own unique

agent_name

so you can dispatch them individually. • Once you set

agent_name

, automatic dispatch is turned off—agents will only join rooms when you explicitly send them via the API. • This setup is super handy for things like telephony integrations or when you have multiple agents and want to control behavior based on the agent type. Sources: Dispatching agents | LiveKit Docs | Making calls using SIP | LiveKit Docs | Agents telephony integration | LiveKit Docs | Accepting incoming calls | LiveKit Docs | Agent dispatch | LiveKit Docs

Can you give me an example of how I can spin up multiple workers with different names like

"my-agent-1"

,

"my-agent-2"

, each running as its own process or service?

Hey there! If you're looking to spin up multiple workers with different agent names in LiveKit, each running as its own process or service, here’s a friendly step-by-step guide to help you out: 1. Deploy Multiple Worker Instances with Unique Agent Names You can start each worker process independently, giving each one its own

agent_name

in the

WorkerOptions

. Here’s a quick example:

from livekit.agents import WorkerOptions, cli

async def entrypoint():
    # Put your agent logic here
    pass

if __name__ == "__main__":
    # Start the worker with a specific agent name
    cli.run_app(WorkerOptions(entrypoint_fnc=entrypoint, agent_name="my-agent-1"))

Then, just run multiple instances of this script, each with a different

agent_name

like so:

# Terminal 1
LIVEKIT_URL=<ws://localhost:7880> LIVEKIT_API_KEY=xxx LIVEKIT_API_SECRET=yyy python agent1.py start --agent_name "my-agent-1"

# Terminal 2
LIVEKIT_URL=<ws://localhost:7880> LIVEKIT_API_KEY=xxx LIVEKIT_API_SECRET=yyy python agent2.py start --agent_name "my-agent-2"

2. Use a Dispatcher to Launch Workers Dynamically If you want to get a bit fancier, you can launch multiple worker processes programmatically with different settings or identities. Here’s how:

import subprocess
import os

def launch_worker(agent_name, room_name):
    cmd = ["python", "agent_runner.py"]
    env = os.environ.copy()
    env["AGENT_NAME"] = agent_name
    env["ROOM_NAME"] = room_name
    process = subprocess.Popen(cmd, env=env)
    return process

# Launch two different agents
launch_worker("my-agent-1", "room1")
launch_worker("my-agent-2", "room2")

3. Production Deployment Best Practices • Run multiple servers or containers, each with its own

agent_name

. • Consider using orchestration tools like Kubernetes or Docker to help manage scaling smoothly. • Each worker will automatically share availability and capacity info with the LiveKit server, which helps with load balancing. If you’d like, I can also share an example Docker setup or Kubernetes deployment config to get you started with this pattern. Just let me know! Sources: Worker lifecycle | LiveKit Docs | Deployment and scaling | LiveKit Docs | Deployment and scaling | LiveKit Docs | WorkerOptions | LiveKit Agents | examples/avatar_agents/audio_wave/dispatcher.py