Self-management

From inside a Sandbox you can edit that Sandbox. The this command, available on every machine, mirrors bdy sandbox but is permanently scoped to the one you are on.

bash
this --help
$

Two things make it different from running the CLI by hand:

• No identifier. The command already knows which Sandbox it is in.
• No credentials. The Sandbox is authorized to act on itself.

That combination is what makes it usable by an AI agent. An agent dropped into a Sandbox does not know the machine's name and has no access token - and still needs to expose the dev server it just started.

What you can do

Run this <command> --help for the exact options of any of them. The help comes straight from bdy sandbox, so every signature still lists the <identifier> argument you leave out here.

Typical uses

Publish what you just started. A dev server is listening on port 5173 and needs to be reachable:

bash
this endpoint add -e 5173
$

Checkpoint before something risky. Save a known-good state before a migration or a dependency upgrade:

bash
this snapshot create -n before-upgrade
$

Keep a process alive. Promote a command started by hand into a supervised app that survives restarts:

bash
this app add "npm run worker"
$

Grow the machine. Raise resources for a heavier build, then bring them back down. There is no dedicated flag for it - this update takes the full YAML definition:

bash
this yaml > /tmp/sandbox.yml    # edit resources, e.g. to 4x8
this update @/tmp/sandbox.yml
$$

Clean up. A task-scoped Sandbox can remove itself once the work is done:

bash
this destroy
$

Knowing where you are

Built-in variables give scripts the Sandbox context - identifier, SSH address, tags, and endpoint URLs. See Variables and secrets.

Next

• CLI
• AI agents