Skip to main content
Your agent runs as a worker - your code, wrapped by zrt.serve(...). The Zero Runtime runtime handles the real-time media (transport, GPUs, STT/LLM/TTS orchestration, and turn-taking); your worker just registers with the runtime and receives sessions. You never operate low-latency media infrastructure. Everything here is driven by the zrt CLI. Install it with pip (Python 3.11+):
See CLI Setup for authentication and next steps.
Deployment architecture: a caller's live session reaches the Zero Runtime runtime, which manages transport, GPUs, and real-time media. The runtime streams the session to your agent worker, deployed on Zero Runtime Cloud via the CLI. The control plane invokes your agent to start the session.

​
How a session flows

  1. A caller starts a live session (from a playground, a web/mobile client, or an inbound phone call).
  2. The session reaches the ZRT Runtime, which Zero Runtime manages: transport, GPUs, and real-time media.
  3. The control plane invokes your agent to start the session.
  4. The runtime streams that session to one of your registered workers, which runs your agent code and responds back through the runtime.

​
Deploying your agent

You deploy your agent to Zero Runtime Cloud with zrt up. Zero Runtime hosts, scales, and runs your agent for you: it pulls your image and schedules the pods on our compute, so you never operate a server. You manage everything (versions, secrets, sessions, and logs) through the zrt CLI.

​
Get started

Install the CLI

Install zrt with pip and authenticate.

Deploy an Agent

Create a project, test it locally, then deploy end to end.

Managing Deployments

Versions, secrets, sessions, and logs.

CLI Reference

Every command and flag in one place.