Cloud platform
What is actually underneath the server you rent.
KVM virtual machines on local NVMe, in 48 regions across four continents, driven by the same API our own control panel uses. Below is how it is put together, including the parts that are not finished.
Network
48 regions, and no charge for talking to yourself.
Deploy where your users are. Region choice affects latency and data residency, not price — every plan costs the same in every location.
Europe
- Frankfurtfra
- Helsinkihel
- Amsterdamams
- Londonlon
- Parispar
- Nurembergnue
- Warsawwaw
- Luxembourglux
- Sinessns
- Madridmad
- Stockholmsto
- Manchesterman
- Milanmil
- Gravelinesgra
- Roubaixrbx
- Strasbourgsbg
- Marseillemrs
- Limburglim
- Eritheri
- Pragueprg
- Viennavie
- Zurichzrh
- Brusselsbru
- Chestercet
Middle East
- Dubaidxb
- Istanbulist
- Tel Avivtlv
Americas
- New Yorknyc
- Ashburnash
- Silicon Valleysfo
- Torontotor
- Vint Hill (VA)vin
- Hillsboro (OR)hil
- Sao Paulogru
- Santiagoscl
- Chicagoord
- Atlantaatl
- Los Angeleslax
- Miamimia
- Seattlesea
- Dallasdfw
- Washingtoniad
- Mexico Citymex
- Beauharnoisbhs
- Denverden
- Palo Altopao
- Honoluluhnl
Asia-Pacific
- Singaporesgp
- Tokyonrt
- Mumbaibom
- Sydneysyd
- Bangaloreblr
- Incheonicn
- Osakaitm
- Hong Konghkg
- Melbournemel
- Tashkenttas
- Seoulsel
- Delhi NCRdel
- Jakartacgk
- Chennaimaa
- Almatyala
- Bakugyd
Africa
- Johannesburgjnb
- One dedicated IPv4, one /64 of IPv6
- The v4 address belongs to the server for its whole life, including through a rebuild. Additional addresses are available on request with a justification, because the pool is finite.
- Private networking inside a region
- Servers in the same region get a second interface on an isolated segment. App-to-database traffic never leaves the region and never counts against your transfer allowance.
- Transfer measured on the way out only
- Inbound traffic is free. Outbound is counted against the plan's allowance — 1 TB on the smallest plan, 6 TB on the largest. Going over does not cut you off; we get in touch.
- Volumetric filtering at the edge
- Layer 3 and 4 flood traffic is dropped before it reaches your hypervisor. Application-layer attacks are yours to handle — a proxy in front is the honest answer there.
Transfer allowances run from 0 TB to 32 TB a month depending on the plan. Traffic between your own servers inside one region is not counted at all.
Compute
Real virtual machines, not containers in costume.
Every server is a KVM virtual machine with its own kernel. You can load kernel modules, run your own firewall, mount what you like and reboot into whatever you built. Nothing on the platform depends on you leaving the machine as we shipped it.
Hosts are AMD EPYC with ECC memory. The exact generation varies by region and over time — if you are sizing a fleet and need to know precisely what you are landing on, ask and we will tell you which host type serves that region today.
The Standard and High frequency lines share physical cores between tenants, which is why they are cheap and why sustained full load is not guaranteed. The Dedicated CPU line does not share: those cores are pinned to your virtual machine and nobody else can take them.
Storage
Local NVMe, and the honest cost of it.
Disks are NVMe drives physically attached to the host running your server. There is no storage network between your database and its data, which is the whole reason small servers here feel quicker than their specification suggests.
The trade-off is real and we would rather say it plainly than bury it: local disks mean a failing host cannot hand your running machine to a healthy one. Recovery is a restore. That is why snapshots are free to take and daily backups are priced at cost rather than sold as a premium feature.
Snapshots and backups
- Snapshots — on demand, free to take
- A full disk image you trigger yourself, typically before an upgrade you are not sure about. Restore in place, or use it as the source for a new server in another region.
- Backups — nightly, 20% of the plan price
- Taken automatically every night, kept for 7 days and stored off the host that runs the server, so a host failure cannot take the backup with it. Switch it on in the configurator or at any time afterwards.
Images
Clean images, and a way to configure them before first boot.
Nothing is preinstalled that you did not ask for — no agents, no telemetry, no vendor packages you have to unpick later.
- Ubuntu 24.04 LTSubuntu-24.04
- Ubuntu 22.04 LTSubuntu-22.04
- Debian 12debian-12
- AlmaLinux 9almalinux-9
- Rocky Linux 9rocky-9
- Ubuntu 24.04 + Dockerdocker
Cloud-init user data is accepted when the server is created, so you can hand over an SSH key, a package list and a first-run script and have the machine ready before you log in. Rebuilding onto a different image takes about a minute and keeps the same IP address.
API
Everything the panel does, it does over the API.
There is no privileged internal path. The control panel is a client of the same REST API you get a token for, which is the only way to be sure it stays complete.
| Method | Path | Does |
|---|---|---|
| POST | /v1/servers | create a server from a plan, region and image |
| GET | /v1/servers/{id} | state, addresses, current resource usage |
| POST | /v1/servers/{id}/resize | move to a larger plan in place |
| POST | /v1/servers/{id}/snapshots | take a full disk image |
| POST | /v1/servers/{id}/rebuild | reinstall from any image, keeping the address |
| DELETE | /v1/servers/{id} | destroy the server and release its resources |
The full reference is still moving, so we have not put it behind a public URL where it would go stale. Ask support for a token and we will send you the current document and tell you when anything in it changes.
Tokens are scoped per project and can be limited to read-only. Requests are rate limited per token, and the limit is returned in the response headers rather than discovered by being cut off.
Limits
What the platform does not do.
Every provider has this list. Most of them keep it internal. Read it before you commit an architecture to us.
A host failure is a restore, not a live migration
Because disks are local NVMe rather than network-attached, we cannot move a running virtual machine off a failing host. If a host dies, your server comes back from its most recent snapshot or backup. This is the single most important trade-off on the platform, and it is why we sell backups at cost rather than as a premium tier.
Regions cannot be changed in place
Moving a server from Frankfurt to Singapore means building a new one and cutting DNS over. Snapshots make the copy quick; the cutover is still yours to schedule.
No GPU instances, no object storage, no managed Kubernetes
We do not sell any of the three. If your architecture needs them, you will need a second provider, and we would rather you knew that now.
Terms are prepaid and not pro-rated on early exit
Destroying a server before its term ends stops future billing but does not refund the remainder. Monthly terms keep that exposure small — that is why monthly is the default.
There is no live status feed yet
The status page is written by hand and dated. Per-region health checks and public incident history are being built.