Burst included.
Pay for the seconds you touch.

We meter compute every second — CPU at $0.000015 per active vCPU-second and RAM at $0.0000045 per working-set GiB-second — and submit idempotent Stripe Meter Events from the control plane after each sandbox stops. Active means CPU-seconds your code actually burned; working-set means memory actually resident, clamped to your committed size. Your monthly base fee on Pro/Team converts 1:1 to usage credit — if you spend less than your base, you don't pay overage; if you spend more, the overage bills at the same per-second rates.

There is no flat hourly rate — cost tracks what you touch. The ceiling for a sandbox that pegs one vCPU and keeps 1 GiB resident for a full hour is $0.054 of CPU plus $0.0162 of RAM, i.e. $0.0702. Most sandboxes bill well under that ceiling: idle CPU meters near zero, and memory the workload stops touching drops out of the working set.

At the same rates as everything else: $0.054 per active vCPU-hour plus $0.0162 per working-set GiB-hour, metered per second, only while the app is running. There is no separate app rate card. CPU is metered on the seconds your app actually burns — not the 8 vCPUs it can burst to — and memory on what your app actually keeps resident, so a typical mostly-idle app bills far below its provisioned size. Idle apps bill almost no CPU, and apps that scale to zero bill nothing while asleep. Builds run inside your app's VM at the same rates — there are no separate build minutes. PR preview deployments bill the same way.

At the same rates as everything else: $0.054 per active vCPU-hour plus $0.0162 per working-set GiB-hour, metered per second while the database is awake — plus storage at $0.15 per GiB-month. There is no separate database rate card and no reservation charge: you pay for the CPU your queries actually burn and the memory your database actually holds resident, never more than the tier you provisioned. Capacity is still guaranteed — database VMs run in a class that is never overcommitted. Auto-suspended databases bill storage only, so a dev database that sleeps most of the day bills just its storage. Point-in-time recovery, backups, and failover are included.

It isn't. We don't charge for outbound transfer on any plan, and inbound is free too. We do expect fair use — egress is metered so we can see it, and if a workspace's traffic is wildly out of line with normal agent, app, and database workloads we'll get in touch rather than bill you by surprise.

Each created microVM counts as one sandbox while it is active. Forks, restored snapshots, and template launches are each metered as their own sandbox from create to stop/delete.

Memory is baked into each template — pick a template with the RAM you need, or bake your own with a different memory spec. CPU works differently: every first-party template gets 8 vCPUs, and because you are billed on active CPU-seconds rather than provisioned cores, there is no smaller CPU size to shop for. On Free, sustained CPU is capped at 2 cores per sandbox — plenty for a code-interpreter cell or an app build, and it keeps one workload from taking a whole host. Pro and Team burst to all 8 with no core cap. Under contention the host fair-shares cores in proportion to each sandbox's vCPUs.

No. Hibernation writes state to disk and stops active vCPU/RAM metering. Storage for hibernated state is currently included at no extra charge.

Every plan includes its persistent-volume quota at no extra charge (Free 1 GiB, Pro 100 GiB, Team 2.5 TiB). Provisioned storage beyond your plan's quota bills at $0.15 per GiB-month, metered continuously and drawn from the same monthly usage credit as compute. Managed database storage counts against the same meter — there is no separate database storage rate.

It is monthly usage credit, equal to about 77 hours of fully active 1 vCPU / 1 GiB usage at current rates. Because idle seconds bill almost nothing, real-world usage usually stretches further. When you exhaust it, Free workspaces are paused: running sandboxes, apps, and databases are stopped and no new ones can be created, until the next month or until you upgrade. Your data is kept, and stopped resources start again the moment you upgrade. Nothing is ever surprise-billed. The workspace simply stays paused — we do not delete apps, databases, volumes or settings for going over the free credit.

Yes — annual prepay on Pro and Team saves about 17% (two months free).

Yes. The core is Apache-2.0. You bring machines, networking, observability, and any cloud costs.

Yes. Monthly plans can cancel at the billing portal. Existing usage for the current period is invoiced normally.