01
Hardware
Ryzen 9 compute with NVMe storage, bought outright and racked by us. Cores are not oversold: when a plan says it gets a core, that core is not also being promised to four other tenants during your busiest hour.
Technology
Six layers, all of them ours.
From the cabinet to the file editor, Shardly runs a stack it owns end to end. This is what that actually consists of.
12 ms
Gateway round trip
Median from the Frankfurt region, probed hourly and published.
99.9%
Platform uptime
Rolling twelve months, excluding customer-caused downtime.
3
Independent regions
A failure in one facility does not take the backup vault with it.
EU
Data residency
Workloads, backups and logs stay inside the European Union.
01The stack
What sits between you and a running process.
Each layer is either hardware we bought or software we wrote. Nothing in this list is a rebadged third-party product.
02
Network
Our own switching in every facility, with a short path from Frankfurt to Discord's European gateway. Round-trip time is probed hourly from inside the region and published, rather than quoted from a launch-day benchmark.
03
Control plane
The dashboard, the in-browser file editor, the console and the deploy path are written here. There is no third-party panel between you and your server, so an edge case becomes a commit instead of a support escalation.
04
Shard supervisor
Liveness is measured from gateway heartbeat acknowledgements rather than process state, so a shard that is up but deaf gets replaced. Restarts are budgeted per shard so a reconnect storm cannot inflict its own rate limit.
05
Storage and backups
Backups are encrypted, versioned, and written to a vault in a different region from the workload they protect. Restores are tested on a schedule, because a backup nobody has restored is a rumour.
06
Identity and access
Passkeys and TOTP, session revocation, and staff access that is scoped and logged. The parts that must never be interesting are deliberately the least interesting parts of the stack.
02Compatibility
Bring whatever you were already using.
Runtimes, databases and tooling the platform speaks natively. Nothing here needs a support ticket to enable.
Node.js
Python
Discord API
PostgreSQL
Redis
Docker
Git deploy
Debian
Bun
NVMe
Let's Encrypt
S3 backups
03Regions
Three German sites, one jurisdiction.
Every region is hardware Shardly bought and racked. We can tell you what is in a cabinet because we put it there.
| Region | Code | Role |
|---|---|---|
| Frankfurt am Main | FRA1 | Compute + gateway edge |
| Falkenstein | FKS1 | Compute + object storage |
| Nuremberg | NBG1 | Compute + backup vault |
Latency figures are measured from inside each region against the live gateway, once an hour. Nothing on this page is a launch-day benchmark.
“If the customer has to tell you your platform is broken, the monitoring was decorative.”
PE
Platform engineeringShardly
Every restart writes a reason to the customer-visible log. Every incident gets a public note on the status page, including the ones nobody noticed. We would rather publish an unflattering timeline than let somebody guess.
Want the detail behind a number?
Everything on this page is measured rather than estimated. If you need the method, the window or the raw figure for a review or a story, ask and we will send it.