zeroruntime

Build real-time voice agents in Python. You describe the agent — instructions, tools, and a pipeline of providers — and the runtime runs it.

Install

pip install zeroruntime

or pin the version:

pip install zeroruntime==0.1.0

Requires Python 3.10 or newer.

Most providers need no extra: their credential is a string you set in the environment, so the base install is enough.

Quickstart

import zeroruntime
from zeroruntime import Agent, Pipeline, Room, function_tool
from zeroruntime.inference import TurnDetector
from zeroruntime.plugins import CartesiaTTS, DeepgramSTT, GoogleLLM, SileroVAD

AGENT_ID = "cascade-basic-agent"


class VoiceAgent(Agent):
    def __init__(self) -> None:
        super().__init__(
            agent_id=AGENT_ID,
            instructions="You are a helpful voice assistant.",
            pipeline=Pipeline(
                stt=DeepgramSTT(),
                llm=GoogleLLM(),
                tts=CartesiaTTS(),
                vad=SileroVAD(),
                turn_detector=TurnDetector(),
            ),
        )

    async def on_enter(self) -> None:
        await self.session.say("Hello, how can I help you today?")

    @function_tool
    async def get_weather(self, latitude: str, longitude: str) -> dict:
        """Fetch the current weather for a location.

        Args:
            latitude: Latitude of the location. Estimate it; do not ask.
            longitude: Longitude of the location. Estimate it; do not ask.
        """
        return {"latitude": latitude, "longitude": longitude, "temperature_c": 28}


def invoke_agent() -> None:
    zeroruntime.invoke(AGENT_ID, room=Room(name="Cascade Agent", playground=True))


if __name__ == "__main__":
    zeroruntime.serve(VoiceAgent, on_ready=invoke_agent)

Run it and the playground URL is printed once, on stdout. Open it to talk to the agent.

Credentials

Set your runtime token in the environment before serving:

export ZERORUNTIME_AUTH_TOKEN="..."

Vendor keys go in the environment too, one per provider you use — each provider names the variable it reads when the key is missing.

Providers

zeroruntime.plugins calls the vendor directly with your key on your account. zeroruntime.inference reaches a subset of the same providers through the runtime gateway instead. The names mirror each other, so swapping the import swaps who makes the call and who is billed:

from zeroruntime.plugins import DeepgramSTT    # your Deepgram account
from zeroruntime.inference import DeepgramSTT  # through the gateway