# Workpaper Instant

Instant gives eligible agent commands a lightweight execution environment and moves to a full Linux VM when a task requires native system capabilities. The logical computer ID stays the same. Eligible idle workspaces can move back to lightweight execution.

Last reviewed: 2026-10-15. [Product overview](https://workpaper.co/index.md).

## Start small, use full Linux when needed

Lightweight execution uses WebAssembly for supported commands. Full Linux escalation is based on command compatibility and native-tool requirements, not simply a percentage of CPU usage. Instant is opt-in; ordinary full-computer workflows remain full Linux.

Supported files under `/workspace`, `/home/user` and `/tmp` transfer between environments. Live process memory, open sessions and background processes do not migrate across that transition; processes start fresh. This differs from pausing/resuming a full VM, which preserves its process state.

## Returning to lightweight execution

Automatic return applies to promoted computers using managed commands. The default eligibility windows require at least one hour since promotion and since the last command needing full Linux, plus at least fifteen minutes without managed-command activity. The illustration on the homepage compresses this lifecycle into seconds.

Changed files must be compatible with the lightweight environment and fit its 64 MiB storage limit. Native binaries, native dependency files, or oversized workspaces keep the computer on the full VM. File changes and deletions are merged back before switching its route; failed merges leave the full computer in place.

If direct VM credentials are issued—for example for a direct shell or connection—the computer stays full because that activity cannot be tracked solely through managed commands. This protection survives restart and stop/start. Older promoted records without connection history also stay full conservatively. Normal idle pause still applies separately.

## Billing and limits

Instant is explicitly opt-in: `workpaper instant my-agent` shows both rates and asks for confirmation. Scripts use `--accept-billing`; SDK creation uses `mode: "instant", acceptBilling: true` in JavaScript or `mode="instant", accept_billing=True` in Python. Ordinary `workpaper run` does not add an Instant fee.

CPU usage costs $0.006/CPU-hour on Pro or $0.020/CPU-hour on PAYG after the included monthly allowance (50 CPU-hours Pro; 10 PAYG). **This CPU fee continues after promotion, plus the active VM's size-based runtime charge.** Small VM runtime is $0.010/hour Pro or $5/150 hours PAYG. Large is $0.020/hour Pro or $5/75 hours PAYG. CPU and VM balances are separate. PAYG preserves the existing 3⅓× pack ratio.

One CPU-hour means one CPU busy for an hour, not one elapsed computer-hour. With two CPU-hours consumed in one small-VM hour, the post-allowance total is $0.022 Pro or approximately $0.073333 PAYG. Promoted CPU is measured by the host, including VM overhead and background work. No CPU usage means no CPU fee; a full VM continues its runtime charge while running, even when idle. Stop it or complete scale-back to end that runtime charge.

A $5 CPU pack buys 250 CPU-hours PAYG or 833⅓ CPU-hours Pro when enabled in billing. Existing purchased CPU quantities remain intact across plan changes. Older Instant computers require explicit acceptance of `instant-cpu-v1` before further execution; no retroactive additive charge is applied. Do not equate the animation with a guaranteed promotion or scale-back time. [Current billing and CLI guide](https://docs.workpaper.co/instant).

[SDK documentation](https://docs.workpaper.co/sdk) · [Published pricing](https://workpaper.co/#pricing) · [Implementation and eligibility rules](https://github.com/ujj-exe/workpaper/tree/instant-move-back/internal/instant) · [Contact](https://workpaper.co/contact)