We are two people who had had enough of buying performance in “player slots” and then guessing what that turns into. So we are building it ourselves: our own platform, our own security framework, billing for what actually runs.
Poko · drag to spin
after “Pom Poko”
1 hour
Shortest rental
We spent a long time looking ourselves, and it was the same every time: one host does not have the game, the next is good but too expensive. You need something for a weekend and pay for a month. Nobody tells you what performance sits behind “10 player slots”, or whether the server will still be as fast tomorrow. In the end you rent a VM from a data centre and set it all up by hand.
That is exactly where we start. No player slots, but real resources with visible utilisation. You pay for the time that runs, from one hour. And behind it are two names rather than a ticket system.
Problem
Fluctuating performance, billing by player slots, too expensive for short use, interchangeable providers.
Approach
Build all of it ourselves: platform, panel, security framework and a billing model without asterisks.
Solution
Fixed resources, visible utilisation, rental from a single hour, and two people you can name.
“A player slot is not performance, it is a number on a sales page. With us the panel says what the machine has and how much of it is yours.”
Aegiron
Named after the aegis, Athena’s shield. Aegiron is our own security framework: eight independent layers across three zones, on a foundation of encrypted disks. Break through one layer and the next one is waiting.
8 layers · 3 zonesFoundation L0
The stack
Traffic runs top to bottom through zones A and B to the workloads. Zone C lies across all of it and watches. L0 is the floor everything stands on.
K3s · multi-clusterCluster-internal
Outer shell
Cloudflare · VPN overlay · UFW · SSH via ProxyJump
NetworkPolicies · fail2ban · deny all by default
NGINX Ingress · cert-manager · client certificates
Linkerd mTLS · identity instead of IP
gRPC interceptors · under 1 ms · over 10,000 req/s
HashiCorp Vault · dynamic credentials · per tenant
Workloads
Over 20 Go microservices · gRPC only internally · multi-tenant
Disk encryption
LUKS on every node · data at rest unreadable
Lies across every layer. All of them report here.
L6Runtime security
Falco + Talon · pod kill through namespace isolation
L7Observability
Prometheus · Loki · Tempo · Grafana
L8Alerting and response
Alertmanager · escalation · automatic bans
Scenario
The real thing, step by step, and what each layer does about it.
Target: containedin under 5 seconds
The road ends immediately. NetworkPolicies only let the compromised service reach what it was allowed to reach anyway.
No borrowed identity. mTLS stops it presenting itself as another service.
Every call is checked. serviceauth-go refuses anything with no policy behind it.
Other secrets stay shut. Vault policies apply per service and per tenant.
The behaviour stands out. Falco spots it, Talon kills the pod or isolates the whole namespace.
All of it is documented. Metrics, logs and traces of the incident are there without gaps.
We get woken. Afterwards. Alertmanager escalates and has L1 block the IP. By then the incident is over.
And at the hardware level: if a disk goes missing or a server is decommissioned, all the data on it stays unreadable.
Cloud Spheres is two people. No investor, no call centre. Whoever answers you also wrote the code for it.
Alan
Founder · engineering & infrastructure
Works on all of it: Aegiron, the nodes, the panel, the backend. Sits deepest in the infrastructure and is the one who gets up at night when a node plays up.
BIB International CollegeFHDW · bachelor completed
Silas
Founder · engineering & panel
Works on just as much of it: panel, backend, automation. Answers the support mail alongside that, and thinks every feature should work without a manual.
BIB International CollegeFHDW · dual-study bachelor, ongoing
BackgroundBoth went through BIB International College to FHDW: one with the bachelor finished, one still on the dual-study track. We do not split the work into engineering and everything else — code, infrastructure and support come from both of us.
As of August 2026
2025
The search
Rented servers ourselves, the same annoyance every time. The idea appears.
Spring 2026
Founded
Company registered, first node rented, Aegiron as a prototype.
Now
Aegiron stands
Eight layers in production, panel being built.
Autumn 2026
Launch
First customer servers, forum from October, first tournaments.
2027
More locations
More nodes across Europe, more games, hourly billing on every plan.
Panel tour without an account, prices without asterisks, rental from one hour.
