Where do agents actually run?

In two places, and both are yours. While you build, agents run on your laptop inside stackbone dev. In production they run in a container you deploy to your own cloud. Stackbone provisions no infrastructure.

The container running your agent is its box. You build the image and bring it up wherever you want, then tell Stackbone where it lives. That record is a deployment: the agent it serves, its public address, and the signing secret that authenticates traffic to it. Register it from the browser, or from a terminal with stackbone link.

The control plane at app.stackbone.ai holds that record and the organization around it. It does not hold your data. Studio and the CLI reach the box directly at the address you registered. A box that only answers inside your own network still works from a machine that can reach it.

Your own network can hold the control plane too. stackbone package writes the agent alone by default, which is the box you register above. --target self-host writes a folder with a second container beside it: a control plane that serves its own Studio. The two share one database and one operator token, and there is nothing to register with app.stackbone.ai.

On your laptop there is no box to register. stackbone dev brings up Postgres, Redis and MinIO with docker compose, boots your agents and workflows in-process, and opens a tunnel so cloud Studio can reach the session on your machine.

Read more

• Connect your box: register a running container from the browser, in three steps.
• stackbone link: the same registration from a terminal or a deploy script.
• Self-host behind a proxy: the two-container folder, and the public addresses it needs.
• Local development: what stackbone dev boots and what it injects.