ZRT: Zero Runtime Python SDK

Serverless voice AI agents for Python. Write the agent in real Python with your own STT, LLM, and TTS providers. Zero Runtime runs the live speech‑to‑speech pipeline (turn detection, denoising, interruptions) at low latency in the cloud.

You write the agent. We run the runtime. No media servers, no GPUs, no autoscaling.

Install

pip install zrt

Requires Python 3.10+.

Quickstart

Get your auth token from app.zeroruntime.ai.

export ZRT_AUTH_TOKEN=<your-token>
export DEEPGRAM_API_KEY=<key>    # speech-to-text
export GOOGLE_API_KEY=<key>      # LLM (Gemini)
export CARTESIA_API_KEY=<key>    # text-to-speech

# agent.py
import zrt
from zrt import Agent, Pipeline, Room, function_tool
from zrt.plugins import CartesiaTTS, DeepgramSTT, GoogleLLM, SileroVAD, TurnDetector

AGENT_ID = "assistant"

class Assistant(Agent):
    def __init__(self) -> None:
        super().__init__(
            name="Assistant",
            agent_id=AGENT_ID,
            instructions=(
                "You are a friendly voice assistant. Keep replies short and natural. "
                "When asked about the weather, call the get_weather tool."
            ),
            pipeline=Pipeline(
                stt=DeepgramSTT(model="nova-2-conversationalai"),
                llm=GoogleLLM(model="gemini-3-flash-preview", thinking_budget=0),
                tts=CartesiaTTS(model="sonic-3.5"),
                vad=SileroVAD(),
                turn_detector=TurnDetector(model="echo-large"),
            ),
        )

    async def on_enter(self) -> None:
        await self.session.say("Hi! I'm your assistant. Ask me about the weather in any city.")

    async def on_exit(self) -> None:
        await self.session.say("Thanks for calling. Goodbye!")

    @function_tool
    async def get_weather(self, city: str) -> dict:
        """Get the current weather for a city.

        Args:
            city: Name of the city to look up.
        """
        return {"city": city, "temperature_c": 28, "condition": "Sunny", "humidity": 55}


def invoke_agent() -> None:
    """Start a session once the agent is registered (fired by serve's on_ready)."""
    zrt.invoke(AGENT_ID, room=Room(playground=True))

if __name__ == "__main__":
    zrt.serve(Assistant, on_ready=invoke_agent)

python agent.py

serve() registers the agent under its agent_id and serves a session to each caller; once it's ready, on_ready fires invoke() to start a playground session and print a link you can open to talk to it.

Pass your Agent class (or any callable that returns a fresh agent) — recommended. serve() then builds a new agent + pipeline per call, so concurrent calls stay isolated.

Concepts

Add a @function_tool method to your agent and the LLM calls it (with your typed args) whenever it needs to. Methods are auto-registered; for tools defined outside the agent, pass them via tools=[...].

Providers

Mix and match. Swap any stage in a line:

• STT: Deepgram, AssemblyAI, Google, Azure, Sarvam, and more.
• LLM: OpenAI, Google Gemini, Anthropic Claude, Groq, Cerebras, xAI Grok, Sarvam, and more.
• TTS: Cartesia, ElevenLabs, Google, Azure, Deepgram, and more.
• Realtime (speech‑to‑speech): OpenAI Realtime, Gemini Live, and more.
• Turn detection: Echo · VAD: Silero · Denoise: RNNoise

Resources

• Docs: https://docs.zeroruntime.ai/
• Examples: https://github.com/ZeroRuntimeAI/zrt-python-sdk-examples
• Dashboard & tokens: https://app.zeroruntime.ai
• Support: support@zeroruntime.ai

© 2026 Zujo Tech Pvt Ltd. All rights reserved.