Service Level Agreement

01 The commitment

Each VPS we sell carries a 99.9% availability commitment per calendar month, measured in UTC. In a 30-day month, 99.9% allows roughly 43 minutes of downtime. Below that line, the credit tiers in §03 apply — automatically.

The commitment applies to every plan, every location, Linux and Windows, from the $5.99 SV-Nano upward. There is no enterprise tier that unlocks the real SLA. This is the real SLA, and everyone is on it.

02 Definitions

SLAs die in undefined terms, so here are the ones we use, precisely:

• Network availability — the datacenter edge serving your location can route traffic to and from the public internet, including our always-on DDoS mitigation layers.
• Hardware availability — the physical node hosting your VPS has power, functioning storage, and functioning chassis components.
• Hypervisor availability — your virtual machine is scheduled by the KVM hypervisor and responds to external probes on its primary IP address.
• Downtime — a minute counts as downtime when your VPS fails probes from all three external monitoring points for that entire minute, due to a network, hardware, or hypervisor failure on our side.
• Monthly price — the monthly-equivalent price of the affected service, add-ons included. On quarterly, semi-annual, or annual cycles, the discounted per-month figure you actually pay.

The measurement window is the calendar month in UTC. Credits are calculated per service, not per account: one affected VPS earns its own credit and leaves the rest of your services out of the math.

03 Credit tiers

Credits post to your account balance, spendable on anything we sell and never expiring. Per the Terms of Service §08, SLA credits are the sole remedy for availability failures — that is the trade that keeps the tiers this simple.

04 Automatic crediting — no ticket, ever

The industry standard for SLA credits is a claim form, a 30-day window, a support agent asking whether you have graphs, and a quiet hope that you’ll forget. We find that model dishonest, so ours works like this:

• Probe data closes out on the 1st of each month at 00:00 UTC.
• Any service below its threshold is credited by the 7th, with a line item in Billing — e.g. SLA credit — June 2026 — atlas (SV-Pro).
• Nothing is required from you. There is no claim process to miss.

You can cross-check us: the trailing 30-day uptime for each service is shown on its manage page in the client area, computed from the same probes we bill against. We grade ourselves with the ruler we hand you.

Why automatic? Because in five years we have watched customers trained by other hosts to accept SLAs that are technically generous and practically unclaimable. Ours pays out on our own data, on the honor system, in public. It has cost us money. That is the point of it.

05 Exclusions

The following do not count as downtime:

• Announced maintenance. Work announced at least 48 hours ahead via the client area and the status page. Most maintenance is handled by live-migration and causes zero downtime; anything that can’t be live-migrated gets a window, posted 48+ hours out, sized to the work and nothing more.
• Customer-caused unavailability. Your OS configuration, your software crashing, your disk filling up — and your firewall: if you drop our probe traffic, those minutes don’t count, because your server was up and simply instructed not to answer. The out-of-band console exists for exactly the day you lock yourself out.
• Your application layer. The VM is up and reachable, but your app is down. We sell the VPS, not what runs on it.
• DDoS targeting customer applications. Attacks aimed at your specific service or content that exceed or evade mitigation — your game server flooded off the map while the network itself carries the attack fine. A failure of our network-level mitigation (1 Tbps+, always on) does count as our downtime; the distinction is where the failure lives. The L7 filter add-on exists for exactly this gap.
• Force majeure. Natural disasters, war, civil unrest, government-ordered shutdowns or seizures, nationwide power or transit failures, sanctions-driven withdrawal of upstream capacity. Rare in five years of operation, listed because pretending these can’t happen is how SLAs become fiction.
• Suspension under the ToS or AUP. A suspended service isn’t down — it’s suspended.

06 How we measure

Three external monitoring points — three different providers, three different continents — probe every primary IP every 60 seconds: ICMP echo plus a TCP handshake on standard service ports. A minute is downtime only when all three points fail simultaneously, so a monitoring node’s own network problem can never burn you.

Probe logs are retained for 13 months. Open a ticket and ask for yours and you get the raw data, not a summary — it is the same dataset the credit automation runs on.

The probe data is the final authority for SLA math. Section §08 covers what happens when your own monitoring disagrees with it in good faith.

07 Historical performance

Fleet-wide monthly average uptime, per year, across all fifteen locations:

The worst single month on record is recent enough that we can describe it exactly: June 2026, Chisinau. A failed NVMe backplane took one node offline for 4 hours 12 minutes before migration completed. That node’s month: 99.42% — inside the 25% tier. Every customer on it was credited automatically on 1 July. The number of tickets opened about it was zero, because none were needed. The full post-mortem is on the status page.

Live and rolling figures per location are always on the status page; this table is the year-level summary.

08 Disputes

If your own monitoring disagrees with ours, open a ticket within 30 days of the credit run and include your data. We compare probe logs line by line. Where the discrepancy is real and on our side — a probe gap, a mis-scoped exclusion — we credit the difference immediately and fix the gap. Running your own monitoring is healthy; half our customers do, and it has caught exactly two probe blind spots since 2021. Both were credited, both were fixed, both customers still host here.

09 Changes

Material changes to this agreement take effect no sooner than 30 days after announcement on the warrant canary page and a site-wide banner — the same mechanism as the Terms of Service. The 99.9% figure and the automatic-crediting rule have not moved since 2021. The direction of every revision so far has been toward the customer, and we intend to keep that streak.