I need boxes that appear and disappear from CI — which of these actually has an API?

Requirement is narrow so I'll state it precisely rather than describe the vibe.

A job in CI needs to bring up a clean box, do a thing, and destroy it. Dozens a week, sometimes several at once. It must be doable from a token in an environment variable with no human in the loop, no dashboard, no email confirmation, no "click the link we sent you". Payment has to work unattended too — I am not topping up a balance by hand every Tuesday.

Half the providers this board likes have no API at all, and the other half have one that exists to resize a disk. Who's actually built for this?

Depends what you're running. How long does each box live and does it need to be the same region every time?

Four to six hours typically, longest maybe a day. Region doesn't matter at all as long as it's consistent within a run.

SporeStack is the one built exactly for this and i think is only one.

Is not a host with API bolted on. Is API with host behind it. No account, no email, no password — you have a token, you pay the token with BTC or XMR or BCH, and then the token buy server. CLI is first class, not afterthought. From 5 dollar for 30 day.

The thing you must understand before you build on it: server have expiry date, not monthly bill. You buy days. When day finish, machine is gone and nobody email you. For your case this is actually perfect because you want it gone anyway, but if you ever put something long-live there you must remember to extend or is deleted, really deleted.

Other problem is granularity. Smallest unit is one day, so your six hour job cost one day. Is fine at your size, is not fine at thousand job.

Servury also deploy fast, 30 second, no email, but API is thinner. ExtraVM have proper API and 50 crypto but the account is normal account with login, so is less clean for what you describe. Anyway SporeStack first, is the answer.

Worth saying out loud since nobody has: SporeStack is a US operator. No account and no email is an excellent property and it is not the same property as jurisdiction.

For ephemeral CI runners that's almost certainly irrelevant — the box exists for six hours and holds nothing. I'd only flag it because people read "no email, pays in Monero" and file the provider under a heading it doesn't belong in, then later put something on it that does matter.

Noted and agreed. Nothing persistent goes near it, that's the whole design.

The token model is genuinely clever and I say that as someone who complains about everything. You prepay a balance that isn't attached to an identity, and the balance is the account. That's how all of this should have worked from the start and it's slightly depressing it took until whenever they launched for someone to build it.

Day granularity at "dozens a week" — has anyone worked out what that actually costs against a normal hourly provider? The privacy is nice but the arithmetic might not be.

Rough arithmetic, since it's a fair challenge.

Thirty runs a week at one day each, cheapest tier, is roughly $5/month worth of days per run if you took the 30-day price literally — but they sell by the day, so call it something in the region of 17 cents a day at that tier. Thirty of those is about $5 a week, $22 a month.

A mainstream hourly provider at six hours a run would be somewhere around a third of that, call it $8. So you're paying maybe fourteen dollars a month for the property of not having an account anywhere.

Whether fourteen dollars is a lot depends entirely on why you wanted no account. If it's convenience, that's an expensive convenience. If it's because a billing relationship is a correlation you're trying not to create, it's the cheapest line in your budget.

Numbers approximate and from the public pricing page rather than my own invoices.

Fourteen dollars is nothing. The dashboard-and-email dance was costing me more than that in irritation alone.

One thing to build in from day one: reconcile what you think is running against what the provider says is running, daily. Ephemeral fleets leak. You will find a box from March still alive in June and you will have no idea what it was for.

Three weeks in, working well enough that I'll write the follow-up now rather than never.

Runs come up in under a minute from the CLI, destroy cleanly, and the token top-up is a monthly thing I do by hand in about ninety seconds. The leak warning above was correct — I already had two orphans in week one, from a job that failed between provision and register. Reaper job now runs nightly and compares the provider's list against my own.

The day-granularity thing has not bitten yet. If the job count triples it will, and at that point the answer is probably to batch the runs onto one box per day rather than change provider.