Before installing
Can I run this on one machine?
Yes, the cluster builder supports a single hybrid node, and it is a reasonable way to evaluate the platform. It is not a reasonable way to run anything you would miss: one node means no redundancy, and a reboot is an outage.
Do I need to know Kubernetes?
Not to install it. Yes, to debug it. The console covers the routine work, but when something is genuinely wrong you will be reading pod events, and no UI removes that.
Does it need internet access after install?
For pulling images and renewing certificates, yes. For its own operation, no: git, CI,
secrets and identity all run inside the cluster. There is a sealed mode for running
genuinely air-gapped, documented in docs/sealed-mode.md.
What does the "alpha" label actually mean here?
That the platform works and is used daily, and that no independent party has reviewed it, breaking changes ship without ceremony, and the maintainer will not be surprised if you find something serious. It is not false modesty and it is not a formality.
Running it
Where are my passwords?
In OpenBao, under secret/platform/<service>. Nothing was written to
disk during install, by design. Read one with
vault kv get -field=<key> secret/platform/<service> or through
the OpenBao UI on the internal tier.
I scaled something to zero and it came back. Why?
ArgoCD self-heal. Git says the replica count is one, so the cluster is put back to one. Declare the change in git, or use the console's power switch, which pauses automated sync and records the previous counts before scaling down.
How do I update the platform?
Pull upstream into your fork, re-run scripts/generate-from-env.sh to
re-substitute your values, and push to Onedev. ArgoCD applies the difference. Because
the committed template holds placeholders and your values live in .env,
that pull is normally conflict-free.
Something is broken and I do not know where to look.
Start with the console overview — node readiness, sync state and endpoint health cover
most of it. Then docs/TROUBLESHOOTING.md, which is written from real
incidents rather than from imagination.
Are backups included?
Longhorn snapshots volumes, and Velero plus MinIO is an optional group for cluster-level backup. Both are off by default and neither is a backup until you have restored from one. Rehearse it.
Getting out again
Am I locked in?
The components are upstream projects with their own data formats. Your application data is in Longhorn volumes and in the apps' own stores; your configuration is plain YAML in a git server you control. Leaving means moving volumes and manifests somewhere else, which is work, but it is not a rescue operation.
How do I remove it?
Delete the virtual machines in Proxmox. The platform provisions VMs and lives entirely inside them; it does not install anything on the hypervisor. Take your data off the volumes first.
Can I fork it and make it mine?
Yes — MIT, and the fork path is a documented workflow rather than an accident. One thing to know: every deployment forwards a sanitised copy of user feedback to a single canonical endpoint. It is a hardcoded constant with no environment override, documented in the infrastructure README, and it is the only such exception in the project.