Skip to main content
State on a computer is ephemeral by default — a destroy tears it down. A checkpoint is the way to make disk state outlive a single computer.

Checkpoints

A checkpoint is a durable snapshot taken without stopping the VM. The computer keeps running while its disk state is captured, so you can mark a known-good baseline mid-task and return to it later.
A checkpoint has no name: its handle is the returned snapshot_id. List them with GET /v1/snapshots (naive computer snapshots, computers.listSnapshots()) and delete one with DELETE /v1/snapshots/{id} (naive computer delete-snapshot, computers.deleteSnapshot(id)) — refused while a live computer was restored from it.

Forks

A fork creates a new computer from a checkpoint. The fork is an independent computer with its own lifecycle — and it is billed like a fresh create: there is no creation fee, just the 10-minute blocks it is billed while it runs.
Use a checkpoint + fork to explore branches from a common setup: prepare a baseline once, then fork several computers to try different approaches in parallel.

Persistent volumes

Not available. A persistent volume separate from the boot disk is not supported: volume on create is refused with 400 validation_failed, and the CLI has no --volume flag. A paused computer already keeps its disk; size it with disk_gb, and take a checkpoint to keep state past a destroy. Promote anything you must keep permanently to Files.

Pause & resume

Pause and resume map directly onto the runtime’s wake/sleep cycle. A paused computer keeps its disk and resumes with state intact — and while paused it bills nothing. Pausing a computer you created does not refund the rest of its 10-minute block already paid, and any exec, fs, snapshot or execStream call on a paused computer resumes it, so billing restarts. This is the same mechanism the durable runtime uses between turns.
CLI

Next: networking

Outbound access, HTTP preview URLs, and the coming egress pinning.