Skip to content

POSTEngineering log

All posts
pricingsandbox0-cloudpersistent-sandboxesai-agents

Sandbox0 Pricing: Pay for Compute, Keep Your State

Written by
Sandbox0 Team
Published

An agent can finish today's work without being finished with its workspace. The repository, installed dependencies, generated files, and intermediate results may all be useful when the next task arrives.

Keeping that workspace should be an easy decision. We have updated Sandbox0 Cloud pricing so that you pay for running compute, while persistent storage and paused sandboxes are free. There is no monthly storage charge.

Each team can retain 2,000 paused sandboxes by default. If your workload needs more, you can request a higher quota for free.

What You Pay For#

Compute costs $0.015 per GiB-hour, billed by the second using the sandbox's configured memory limit and running time. The hourly rate makes the price easy to compare; it does not impose a one-hour minimum. CPU is included in that rate, with CPU limits still applying.

ResourceCurrent price
Running compute$0.015 per GiB-hour, billed by the second
Persistent RootFSFree
Snapshots and retained memory checkpointsFree
Paused sandboxesFree
Network ingressFree
Network egressFree for now
Object-store GET and PUT requestsFree

Running time includes time when the sandbox is idle or waiting for input. Billing follows the configured memory limit, rather than the amount of memory a process happens to use. Pause the sandbox when you want compute charges to stop. They begin accruing again when it resumes running.

The pricing page lists the current rates and billing boundaries.

A Workspace You Can Come Back To#

Consider a sandbox configured with 4 GiB of memory. It runs for a total of 10 hours, across one session or several, and retains 20 GiB of persistent state. You pause it between tasks.

Its compute charge is:

text
4 GiB × 10 running hours × $0.015/GiB-hour = $0.60

The retained storage adds $0.00. The paused time adds $0.00. Network traffic and object-store requests add $0.00 at the current rates. The example total is $0.60, with no monthly plan fee added.

You can keep the workspace ready for the next task without paying for the time it spends paused. When you resume it, the next running interval adds to your compute usage.

Why a Paused Sandbox Quota?#

Persistent state still has a cost to operate. Previously, we passed storage costs through as a separate charge. We are now absorbing that cost and using a paused sandbox quota to manage retained capacity.

The default is 2,000 paused sandboxes per team, separate from the team's active sandbox capacity. Paused sandboxes do not occupy active capacity. The paused quota counts sandboxes; it is not a bundle of billable storage or an allowance that turns into an overage fee.

At the paused quota limit:

  • New sandbox creation and forks are blocked.
  • Existing running sandboxes can still pause, even if that temporarily takes the team above its paused quota.
  • Existing paused sandboxes can still be resumed or deleted. Resuming remains subject to active capacity and available compute credit.

Resuming or deleting paused sandboxes frees paused capacity. We do not automatically delete your sandboxes for reaching this quota, and going above it does not introduce a storage charge.

If you need more retained workspaces, email [email protected] to request a free quota increase. Include your team and the capacity you need. Increases are subject to approval; they do not require moving to a paid storage tier. You can check your team's effective quotas in the Dashboard.

Built for Work That Continues Later#

An agent's useful lifetime often extends beyond a single run. It may return to a project tomorrow, wait for a person to review a change, or reuse an environment prepared for a particular customer.

Our pricing now follows that rhythm: running time contributes to the bill; paused time and retained state do not. This update is live for Sandbox0 Cloud. Usage from before the change retains its historical rates.

Start building, or visit the pricing page to plan your workload.