service agreement

a single node cannot honestly promise four nines. that is 52 minutes of downtime a year, and one kernel panic spends the whole budget. any small host quoting it is either copying a template or counting only the outages they feel like counting.

• your service is down: we start working within 1 hour, any time of day.
• something is broken but running: a person replies within 3 working hours.
• a question or a request: within one working day.
• working hours are 09:00–19:00 CET, monday to friday. an outage is not held to working hours.

the clock starts when your ticket lands in the panel. the reply is from a person who read it — not an autoresponder assigning a number.

• we miss the reply window: that month is free. all of it, not a prorated slice.
• your machine is down more than 8 hours in a month through our fault: that month is free.
• we lose a snapshot you took and cannot produce it: everything you have ever paid us, back.

you claim it by writing to us. we do not require a form, a deadline, or an explanation — the ticket timestamps are in the panel and we can read them ourselves.

being honest about the exclusions is the only way the promises above mean anything:

• your own configuration, your own software, your own full disk.
• downtime you caused by not paying — that path is described in the terms, with three days of grace.
• the datacentre losing power or connectivity. we pass through whatever compensation we receive for it and we tell you what happened.
• an attack aimed at you that exceeds upstream filtering capacity.

ours — the panel, dns, the things we run — is announced at least 72 hours ahead and scheduled between 02:00 and 06:00 CET.

the machines themselves run on another provider's cloud while we build our own, and their maintenance is theirs to schedule. we pass on what they announce the moment they announce it, and we do not pretend to control the calendar.

announced maintenance inside those limits does not count against the downtime figure above. anything unannounced does.

• you take a snapshot of the disk yourself, from the machine page, whenever you want. it takes about a minute and costs nothing.
• there is no automatic daily backup and no off-site copy. we are not selling one until it exists — a backup you have been charged for and never seen is worse than none.
• a snapshot sits with the same provider as the machine it came from. that covers your own mistakes, not that provider disappearing.
• rolling back onto a snapshot is a button on the machine page — you do not need us for it, and you do not wait for us to be awake.
• keep your own copy of anything you cannot lose.

after an outage of our making we write what happened, why, and what we changed so it does not repeat — in plain language, without «a service degradation was observed». if the cause was stupid, the note says it was stupid.