- 99.9% availability per server per calendar month, measured by our probes from two regions, published on the status page.
- Every minute beyond the allowance is credited ten times over, to your balance, automatically, without a ticket.
- Hardware faults are repaired from local spares with a four-hour target; if we cannot within 24 hours, the rest of the term is refunded.
1. The commitment
Each server is available at least 99.9% of the minutes in a calendar month. For a 30-day month that allows about 43 minutes of downtime. Availability is committed per server, not averaged across a fleet or a region.
2. How availability is measured
Our probes, located in two different regions, test every server every 60 seconds: an ICMP echo to its public IPv4 and a TCP handshake on the SSH port (or the RDP port on Windows). A server is down from the first minute in which both probes fail two consecutive checks, until the first minute in which one probe succeeds. Hardware faults that make the server unusable (a failed card reported by the driver, a failed NVMe holding the system) count as downtime from the moment we detect them.
The same measurements produce the 90-day figures on the status page. We do not rely on your report to start the clock, though you can open a ticket if you believe we missed an outage.
3. What does not count
- Anything inside the operating system: a crashed kernel, a firewall rule that blocks the probe, a full disk, a process that stopped answering.
- Your own actions: reboots, re-images, cancellations, suspension for non-payment or for a breach of the acceptable-use policy.
- Scheduled maintenance announced at least 72 hours ahead, up to four hours per month.
- Emergency security maintenance that could not reasonably wait.
- Events beyond our network: an upstream Internet outage beyond our transit providers, a volumetric attack against your server, force majeure.
4. Credits
For every full minute of downtime beyond the monthly allowance, we credit ten times the pro-rata monthly rent of that server for that minute. Example: a server rented at $1,134 a month, down for four hours in a month, has 197 minutes beyond the allowance; the credit is 197 × (1,134 ÷ 43,200) × 10 ≈ $517. Credits are capped at 100 % of that month's rent for the server, are added to your balance within 24 hours of the end of the month, and appear in the ledger as a line of their own. No ticket is needed.
5. Hardware repair
A failed GPU, NVMe device, memory module or power supply is replaced from spares held in the same data centre, with a target of four hours from detection. The spares reserve is a line of the cost sheet. If we cannot restore the server within 24 hours, you may cancel it and the remainder of the current term is refunded to your balance in full.
6. Provisioning
A single-card server is online within under 5 minutes of the order, an eight-card node within fifteen minutes. If a server is not online 24 hours after the order, you may cancel it for a full refund of that order.
7. Reporting
Incidents, their timeline and their resolution are published on the status page and available as JSON at /status.json. Post-incident notes are added within five working days for any incident longer than 30 minutes.