Skip to main content
This guide deploys Hello Agent, the smallest agent that follows the AgentOS contract: one capability, echo, that returns its input. You need an AgentOS account with the admin role in an environment, because deploying is an admin action. You don’t need Docker on your machine: AgentOS builds the image.
1

Write the agent

Create a folder hello-agent/ with three files.
agentos.yaml names the agent, the port it listens on, and its capabilities with their input and output schemas. See agentos.yaml for every key.
2

Zip it

The Dockerfile and agentos.yaml must be at the top level of the zip, not inside a folder:
3

Deploy it

Open Agents → Deploy agent, choose hello-agent.zip and click Build v1. The build log streams on the page. When the container answers /health, the version becomes v1 and the agent is live.
4

Run it

Open the agent’s Run tab, pick echo, enter {"message": "hi"} and run it. The run page shows its input, output, timeline and duration.
5

Ship a change

Edit the agent, zip it again and upload it on the agent’s Deploy tab (Deploy live) or to the same API URL. It becomes v2 once it is healthy, and v1 stops but keeps its image: roll back from the agent’s Deploy tab at any time.

Next steps

Call a model

Use the environment’s models through the gateway with the run token.

Ask a person

Pause a run for an approval and continue with the decision.

Use an SDK

The same agent in a few lines of Python or TypeScript.

Examples

Working agents for models, approvals, streaming, memory and frameworks.