Your own software
A kernel panic from a module you loaded, a full disk, a firewall rule that locked you out. We will help you fix it; it does not count against the SLA.
99.9% committed. 99.894% actually delivered.
The commitment is the floor at which we start owing you money. The second figure is what the last ninety days measured, published on the same page so the two can be compared.
- Committed
- 99.9% monthly
- Last 90 days
- 99.894%
- Last 365 days
- 99.954%
- Claims required
- None
The bands
What a bad month costs us
Per service and per calendar month, applied to the next invoice without you doing anything. The minutes column is what each band means in practice, because a percentage of a month is not a quantity anyone can picture.
| Availability | Downtime | Credit | What this band usually means |
|---|---|---|---|
| 99.9% – 99.5% | Under 44 min | 10% | A single reboot that overruns, or a switch that took longer than it should. |
| 99.5% – 99.0% | 44 min – 3 h 39 | 25% | A fault we did not catch quickly enough. This is the band where we owe an explanation, not just a credit. |
| 99.0% – 95.0% | 3 h 39 – 36 h | 50% | Something structural failed. Expect a written post-mortem without asking for one. |
| Below 95.0% | Over 36 h | 100% | The month is free. There is no scenario in which we argue about it. |
Measured per service, not as a network average — an average lets a provider hide one customer’s terrible month inside everyone else’s good one. Actual availability over the last 365 days was 99.954%, and the status page shows the incidents that produced it.
The part that decides everything
Five exclusions. This is all of them.
Most service level agreements die here, in a clause nobody reads. Ours is five items long and “force majeure” is defined rather than left to mean whatever is convenient on the day.
Scheduled maintenance
Announced at least 72 hours ahead on the status page, capped at 4 hours per quarter, and almost always live-migrated instead.
Suspension for the two refusals
Child sexual abuse material or material organising violence. Removal is immediate and no credit applies.
Non-payment
An invoice unpaid past its grace period. The service stops for a reason you control.
Force majeure
War, natural disaster, or a national-scale outage of both carriers at once. Narrowly defined on purpose — “force majeure” is the clause every provider hides behind.
Not excluded, and worth naming: hardware failure, a failed upgrade, a configuration mistake, a routing error, a denial-of-service attack, or a datacentre power event. Every one of those is an ordinary operational risk, and absorbing it is what you are paying for.
No claim to file
A credit you have to chase is not a credit
Four properties that decide whether a service level agreement is a commitment or a procedure designed to be abandoned halfway through.
You never file a claim
Downtime is measured on our side and the credit is applied to your next invoice by the same job that measured it. There is no form, no window in which to submit, and no evidence to attach.
No proof of impact required
Providers that ask you to demonstrate business loss are asking for something nobody can produce in the time allowed. If the component was down, the credit is owed.
Applied within one billing cycle
Not held over, not offset against a future increase, not converted into service extension you did not ask for. It comes off the next invoice.
Refundable, not just creditable
If you leave with a credit balance, it is returned in the coin you paid with rather than quietly expiring. This is unusual and it is deliberate.
The document
The operative text. It forms part of the terms of service, and where the two differ on an uptime question, this document governs.
What is committed
Each service is available at least 99.9% of each calendar month, measured per service rather than as a network average. A month with 43 minutes of unplanned unavailability meets the commitment; a month with 45 does not, and the credit applies to the affected service alone.
Availability means the service responds on the network. A machine that is up but unreachable is unavailable, because the distinction matters to you and not to us.
How it is measured
By probes on three networks that are not ours — Frankfurt, Amsterdam and New York — at 60-second intervals. A service is counted down when at least two of the three agree, which stops a single probe’s own connectivity from generating credits nobody experienced.
Downtime is counted from the first failed check, not from when a ticket was opened, and it runs until two probes agree the service is back. Both ends of that definition cost us minutes rather than saving them.
Credits
Applied automatically to the next invoice, as a percentage of the monthly fee for the affected service, according to the bands published above. Credits are cumulative across services but capped at 100% of the monthly fee for any single service in any single month.
You do not submit a claim. There is no notification window, no evidence requirement and no discretion on our side — the same job that records the downtime applies the credit.
If a credit you expected does not appear, that is a bug in our billing rather than a dispute about your entitlement. Tell support and it is corrected, backdated.
What is excluded
Five things, and they are the whole list: faults in software you installed; scheduled maintenance announced at least 72 hours ahead and capped at four hours per quarter; suspension under the two content refusals in the acceptable use policy; non-payment past the grace period; and force majeure.
Force majeure is defined narrowly on purpose: war, natural disaster, or the simultaneous failure of both transit carriers for reasons outside Moldova. It does not cover a hardware fault, a failed upgrade, a configuration error, a routing mistake, or a denial-of-service attack — all of which are ordinary operational risks that we are paid to absorb.
Scheduled maintenance
Announced on the status page and by email at least 72 hours in advance, with a stated window. Most of it is live-migrated and you see nothing; where a reboot is unavoidable, affected customers are told individually rather than by a general notice.
Maintenance that overruns its announced window stops being maintenance at the moment the window closes, and the overrun counts as unplanned downtime.
Network commitments
Packet loss inside our network is committed below 0.1% and latency between our edge and either transit carrier below 2 ms. Sustained breach of either is treated as unavailability for the affected service, at the same credit bands.
We make no commitment about the internet beyond our transit carriers, because nobody can. What we do commit to is telling you which side of that boundary a problem is on, with the traceroute, rather than asserting it is yours.
Post-mortems
Any single event over one hour, or any month falling below 99.5%, gets a written post-mortem published on the status page within seven days: what happened, what the cause was, and what changed. It is published whether or not anyone asked, and whether or not it reflects well on us.
Persistent failure
If a service falls below 99.0% in two consecutive months, you may terminate it immediately with a full refund of the current term, and the credits already applied are not set off against that refund. No notice period applies and no exit fee exists.
Questions about a specific month are raised from the customer panel, and are answered with the probe data rather than with a summary of it.
Version 1.3 · last substantive change 2026-07-01 · in force since 2026-07-22. Material changes are announced by email 30 days before they take effect, and the previous version stays reachable.
The number behind the promise
What we committed, next to what we delivered
Both figures live on the status page, with every incident that moved the second one.