Skip to content

Sandboxes and terminals in the console

What you can do to a sandbox from an opened agent on the Fleet. Then how the terminal drawer works, and what Hide, Destroy and Snapshot keep.

Sandbox actions

On a running sandbox:

  • Terminal: a shell;
  • Run agent: a headless agent on a task, in its own terminal, which opens;
  • Run command;
  • Fork: snapshot it and start copies with the same files, dependencies and repo state;
  • Snapshot, Pause and +1 h on its lease.

A paused one offers Resume and +1 h. Every sandbox that isn't stopped has Destroy. Open job or Open session goes to the work's own page.

Terminals

Terminals open in a drawer at the bottom of the console, one tab each; drag its top edge to resize it. Opening an agent run attaches to the same terminal the agent uses: you see what it types and can type back.

Terminal opens a plain shell. To switch to a terminal an agent or an MCP client is using, choose Sessions…, pick it and Attach. A closed connection reconnects when you press Enter. Escape inside a terminal goes to the program running there (vim, less), not to the console.

Hide, Destroy and Snapshot

  • Hide, in the terminal drawer, only hides the drawer. The terminals stay open and the sandbox keeps running. Close a terminal with its tab's ×.
  • Destroy, in an opened agent, deletes the sandbox and its filesystem, after you confirm, and closes its terminals. Push or save what you want to keep first. A job's branch and a session's conversation are kept. A session's sandbox is checkpointed before it goes, so the session suspends and resumes later from that checkpoint.
  • Snapshot keeps the sandbox's state without stopping it: a full snapshot where the provider has them, otherwise the workspace's files. Snapshots are listed under Snapshots, where Start copy starts a new sandbox from one. Take one before you destroy a sandbox whose files you may need.

From the CLI, hangar sb rm <name> removes a sandbox.