Skip to main content
Sandbox data deletion and retention depend on the resource. Pause preserves machine state; deletion retires the resource through its lifecycle operation.

​
How long data stays

​
Paused storage and credit grace

Paused persistent machines incur storage charges. If funding cannot renew, the default grace period is 14 days; the period is configurable. Inspect the returned grace deadline rather than calculating a deadline from this page. Memory and disk remain during grace. Storage that cannot be charged during grace is a platform cost, with no retroactive storage debt after top-up. Adding credit can fund a new window. After grace ends, normal lifecycle deletion retires the machine. Snapshots use the same funded storage and credit grace model. A caller-set snapshot expiry is independent of funding. See Snapshot retention.

​
Delete a machine

Check the returned deletion operation and wait for completion when the API returns an operation ID. Its status is available at GET /operations/OPERATION_ID:
A submitted delete or missing response does not by itself confirm completed cleanup. Read status before retrying an uncertain operation. Confirmed authorized usage can still settle after deletion.

​
Clean up separate resources

Delete saved snapshots and independent fork children separately. bead worktree down removes the bound machine; prepared snapshots and templates remain separately retained. Volumes have independent retention deadlines and remain after machine deletion releases their attachment during cleanup. Detach an attached volume before requesting DELETE /volumes/VOLUME_ID, then inspect GET /volumes until cleanup completes. Remove a project vault name with bead env unset NAME from the project. Local source caches remain until you remove them; acknowledged worktree down removes that machine’s local artifact cache, so copy needed artifacts elsewhere first. Deleting a local secret does not erase its contents from an existing remote snapshot.