Regions and trust
Where do boxes run? In the EU: Germany, Finland, and France. Your data and snapshots stay there. Is EU latency a problem from the US or Latin America? Round trips are around 100 to 200 ms. Agentic work runs for minutes, so this is not noticeable in practice. For end-user production traffic, publish builds to a CDN or regional host and keep Box for the agent and dev preview. See Build a Platform on Box. Do you have compliance certifications? Not yet. SOC 2 is in progress. Boxes are full VMs with sudo, so you can add any controls you need inside them (firewall rules, proxies, encryption).Billing
Am I billed when a box is stopped? No. Stopped boxes are free, and their snapshot is kept for the life of the box. What do snapshots, IPv4, and egress cost? Nothing extra. The latest snapshot per box is kept (up to your box type’s data size: 12 GB onsmall, 50 GB on default, 70 GB on large), every box has a dedicated IPv4, and egress is included up to 2 TB per box per month. See Billing & Limits.
Do unused credits expire?
Plan time resets monthly. Purchased credit packs never expire.
Lifecycle
How do I get rid of a box? Almost alwaysbox stop: it snapshots, is free and reversible, and you can resume any time, months later included. Stopped boxes cost nothing, so there is rarely a reason to delete one.
If you do want it gone, box delete <id> (or Delete in the dashboard ⋯ menu) force-stops the box and permanently deletes its snapshots. It cannot be resumed or forked afterwards and there is no undo, so reach for it only when you are sure the data is worthless. Snapshot data another box forked or resumed from it still uses is kept, as are named snapshots saved from it. See Snapshots.
Why does a box sometimes come back up instead of stopping?
Stopping saves the disk first. If the box has no successful snapshot in the last 30 minutes we take one to prove the pipeline works, and if that fails we leave the box running rather than throw away your work. You are not billed for that time, the meter pauses from the first failed attempt. If it persists, snapshots are failing on that box: contact us. We are alerted automatically, fix these the same day, and refund the extra runtime. To stop anyway and accept losing everything written since the last successful snapshot, pass force.
How fast are resume and fork?
A few seconds, whatever the box holds. Files stream in the background for the first moments after access is given; reading a file that has not arrived yet simply waits until it has, it never returns wrong data. Wait for state ready or idle before running commands; commands sent earlier run before your environment variables are applied.
Do processes survive stop and resume?
Files, installed packages, and enabled systemd services do. Hand-run processes do not; restart them or make them a systemd service. See Snapshots.
Running a platform
One box per user, per project, or per session? Compare the three patterns and their costs in Build a Platform on Box. How long should I leave a user’s box running before stopping it? Box has no idle timer. The auto-stop TTL counts from creation or resume, never from last activity, so run your own countdown and callbox stop a few seconds to a couple of minutes after the agent finishes. Resume takes seconds, so a short countdown cuts cost with little UX impact. Budget the resumes: each one counts against your machine start limits exactly like a create.
What counts as a machine start, and what are the limits?
Create, fork and resume each count as one machine start. Your plan sets the per-minute rate, and every account also sits under platform ceilings of 600 starts per hour and 1,500 per day. So a spike can run hot for an hour, while the day as a whole is bounded at about two and a half such hours. If you need more, contact us. See Billing & Limits.
How do I handle launch spikes?
Prepare a template box and box fork it per user instead of installing from scratch, and check that your start limits cover the peak. See Template Boxes.
Do prepared templates actually reduce billed time?
Yes. A fork inherits the whole filesystem in seconds, so you never pay for repeated installs.
Is Box good for sustained heavy compute?
Boxes have shared vCPUs and shine at agentic, IO-heavy, spiky work. box new --type large doubles the CPU and memory of a default box if a workload needs more headroom. For sustained multi-hour 100% CPU or GPU loads, dedicated compute primitives are a better fit.
Can I resize a box?
Yes. Pass --type to resume or fork to move a box between small, default and large. Growing always works; shrinking is refused when the box holds more data than the smaller machine takes, and the box is left exactly as it was. bare-metal is the one exception, it cannot convert to or from the cloud types. See Machine Capabilities.