Skip to content
Alpha: for testing only; it does not yet meet the Cyber Resilience Act’s requirements. Status and limits →

Environments and projects

A hub manages servers of three kinds, each running an environment (your YAML from Git, run by Podman):

  • production: what your users reach;
  • staging: a server that gets releases before production;
  • twin: a short-lived copy of a production or staging environment for one task. See twins.

A hub knows two cloud projects, as two fixed settings:

  • the hub’s own project: the hub, production and staging, on one private network;
  • the twins’ project: twins, recovery tests and, later, agent servers.

The production token never reaches a twin. A Hetzner Cloud token covers its whole project, so a separate twins project limits what a leaked twin token reaches.

factconsequence
a token covers its whole project, read and writemore projects limit what a leaked app or cluster token reaches
a private network belongs to one projectservers in different projects talk over public addresses
Cloud Firewalls do not filter traffic inside a private network (Hetzner’s firewall FAQ)everything on a private network can reach everything else on it
quotas are per projecta busy twins project cannot eat production’s quota
projects cost nothing and show separately on the invoiceseparation by project is free

The firewall row is the subject of a blog post.

Decided in part on 23 October 2026, the rest proposed: a project becomes an object in the hub with a token, a purpose and its networks, and a hub holds as many as it needs.

  • Five purposes: management (the hub), production, nonproduction (staging and the like), twins, and sandbox (agent servers), later a sixth, infrastructure.
  • Two hard rules: a token of a production, management or nonproduction project never enters a twin or a sandbox; nothing of purpose twins or sandbox joins a private network that production uses. The hub refuses a binding that breaks them.
  • Agent sandboxes join no private network at all. An agent server reaches its own twin only through the hub.
  • Layouts: small (two projects, as today, what vai up makes), business (one project per purpose, for teams with an auditor), agency (per client).

It is scheduled after the CLI over HTTPS.