Sandbox

Configure a sandbox provider on the agent in the dashboard to enable files, commands, and published results. Each conversation gets an isolated Linux sandbox, with no additional SDK wiring.

In the chat

• Files the user attaches are written to uploads/ in the sandbox, so the agent can analyze a spreadsheet with real code. See Attachments.
• Each sandbox step is a transcript row: a file write expands to the file, a command to its output.
• While the sandbox provisions (tens of seconds), a slim status pill sits above the composer.
• Generated images (PNG, JPEG, GIF, WebP) the agent publishes render inline, with a download button.
• Any other published file appears as a download row. Downloads are authorized by short-lived tickets.

The sandbox panel

Off by default: the transcript already shows each step where it happened. sandboxPanel: true opts in.

<AstralBeamChat sandboxPanel />

• A pill above the composer summarizes the sandbox ("3 files · 5 commands") and toggles the panel.
• The panel opens as an anchored sheet, so the composer never moves and a streaming reply stays readable.
• Files lists the latest content of every file the agent wrote, each viewable and downloadable.
• Log is the whole command history with exit codes and durations. Failures read in red.

Limits

• File and command output shown in the chat is clamped server-side for model context.
• Published artifacts are capped at 10 MB and served with content-sniffed types. See Security model.
• The sandbox is reused across turns of one conversation and expires after ~15 idle minutes. Expired downloads say to ask the agent to regenerate.