openrtc 0.0.6

Run multiple LiveKit voice agents in a single shared worker process.

OpenRTC

OpenRTC is a Python framework for running multiple LiveKit voice agents in a single worker process with shared prewarmed models.

Decorator-based discovery with shared defaults

Use @agent_config(...) on a standard LiveKit Agent subclass to attach optional discovery metadata. Then set shared providers once on AgentPool and only override the values that differ per agent.

from pathlib import Path

from openrtc import AgentPool, agent_config
from livekit.agents import Agent


@agent_config(name="restaurant")
class RestaurantAgent(Agent):
    ...


pool = AgentPool(
    default_stt="deepgram/nova-3:multi",
    default_llm="openai/gpt-4.1-mini",
    default_tts="cartesia/sonic-3",
)
pool.discover(Path("./agents"))
pool.run()

The CLI also accepts shared defaults for discovered agents:

openrtc list \
  --agents-dir ./examples/agents \
  --default-stt deepgram/nova-3:multi \
  --default-llm openai/gpt-4.1-mini \
  --default-tts cartesia/sonic-3

For backward compatibility, discovery still supports legacy module-level AGENT_* variables, but the decorator is the preferred pattern.

Session routing and greetings

AgentPool resolves sessions in this order:

1. ctx.job.metadata["agent"] or ctx.job.metadata["demo"]
2. ctx.room.metadata["agent"] or ctx.room.metadata["demo"]
3. room name prefix matching such as restaurant-call-123
4. the first registered agent

Registered greetings are generated after ctx.connect(), and advanced AgentSession options can be passed either with session_kwargs or directly as keyword arguments to add(). Direct keyword arguments take precedence when the same option is provided both ways.

pool.add(
    "restaurant",
    RestaurantAgent,
    greeting="Welcome to reservations.",
    session_kwargs={"allow_interruptions": False},
    max_tool_steps=4,
    preemptive_generation=True,
)