Region
Sandboxes run in a single region today. The API, the dashboard and live view are served from the same region. Coming soon: more regions.Plan limits
Prices are on Pricing.
- At once means running. Sandboxes that are being created or are running count towards “sandboxes at once” and “vCPU at once”; paused sandboxes never do.
- A sandbox created without a size runs, on Free, at the smallest size its template allows: Small (1 vCPU, 2 GiB) for
base,python,node,code-interpreterand templates you built at Small or less, so five run at once; Medium fordesktopandagent-desktop, which count towards the two Medium. On paid plans it runs at the template’s own size and counts at that size. See Sandbox sizes. - The largest size applies to every create. A size above it is refused, and so is a sandbox created without a size from a template registered above it. A sandbox you already have keeps its size if your plan changes.
- API requests counts requests made with an API key. The dashboard’s own requests do not count.
- Template storage is the sum of the image size of each template’s current build. The build before it is kept so a rebuild can be rolled back, but it is not counted. One template image is at most the largest template image size. See Templates.
- Paused sandboxes on Free are deleted 14 days after they were paused. We email the team’s owners and admins three days before. Resuming the sandbox starts the 14 days again the next time it is paused.
- Included vCPU-hours and egress are per calendar month (UTC) and reset on the first. Each included vCPU-hour includes 2 GiB-hours of memory. Free stops when either runs out: new sandboxes and resumes are refused, and running sandboxes are paused (or deleted, end reason
included_usage, when your paused sandboxes are already at the limit) until the month ends. Paid plans keep running and are billed for what is over. A credit balance you already have pays for what is over first, on any plan; Free stops only once it is spent. Owners and admins are emailed at 80% and 100% of each. Usage is counted hourly, so a team can run past the line by up to about an hour before it applies. See Pricing. - Public internet egress is on by default on paid plans. On Free it is off for now (see the table): the 5 GB a month needs a cap that acts in real time, and usage is counted hourly. Coming soon: Free egress on, capped at 5 GB.
- Vault secrets have no limit on any plan.
When you reach a limit
Nothing you already have is stopped or deleted because you reached a limit, or because you moved to a plan with lower limits. New sandboxes, resumes, pauses, API keys and templates are refused until you are back under it. The one exception is Free’s included usage for the month: when it runs out, running sandboxes are paused too. Every refusal names the limit and your plan. A refusal at a plan limit haserror_code plan_limit_exceeded and four more fields:
Other limits
These apply as well and are not part of the plan table:
Every plan allows
timeout=-1, which means no deadline; see Lifecycle. Free running time still counts against the month’s included vCPU-hours. Across every plan, a team is also limited to about 50 requests a second (bursts of 200) and 120 creates a minute; above that the answer is 429 rate_limited with Retry-After. Builds in progress count queued builds too: the shared builder runs one build at a time today, so a second build waits in the queue (its log says so) until the first finishes.