Skip to main content
Use memory snapshots when you want to reuse prepared process state across later machines. Keep the snapshot ID and wait for readiness before restoring work.

​
Save a snapshot

The API is POST /sandboxes/{id}/snapshots. The response is 202 pending with snapshotID, templateID, and buildID. Keep the snapshot ID.

​
Wait for readiness

Poll until status is ready. A pending capture is not ready for restored execution. If capture fails, inspect the source machine’s preservation object before deciding whether to retry. If snapshot detail reports retryable: true and nextAction: "retry_or_retire", retry the existing capture with bead api POST /snapshots/SNAPSHOT_ID/retry, then poll its status again. Retire it instead if you no longer need it.

​
Snapshot restore

Use the snapshot’s original size in place of small when different. Memory restores require the original CPU and memory bounds. Check restored readiness before sending work.

​
What is copied

The snapshot captures process memory and the machine’s disk and devices. Tenant environment values and secrets already in memory can be retained. Credential-profile inheritance requires explicit permission; a credential-bearing source can refuse capture when consent is missing.
For runner-prompt credentials, authorize capture with --allow-credential-snapshot; for a create-time profile, set metadata["bead.allowCredentialSnapshot"] to "true" in the create API request. Capture with bead sandbox snapshot create ID --name prepared --inherit-credentials, then separately set metadata["bead.inheritCredentials"] to "true" in each restore’s create API request.
Network connections do not migrate transparently. External peers may require reconnecting, and applications bound to an old private address may need to rebind. A restored machine gets its own platform identity. Saved snapshot capture requires a source with no attached persistent volume. Close files and detach the volume before capture; volume contents and retention remain separate from the snapshot.

​
Keep or retire a snapshot

Snapshots have no scheduled expiry by default while storage remains funded. Optional expiry and storage grace still apply. See snapshot retention, fork, and pause and resume.