Selfhosted Cloud

your cloud OS

affineblockscoutciso-assistantclickhousecrowdsecdefectdojodefectdojo-classicdependency-trackdocusealethereumexternal-imapexternal-mongodbexternal-mysqlexternal-oidcexternal-postgresexternal-redisexternal-smtpfleetfrappefusionpbxgiteagitlabgitlab-runnergraphitigreenbonehighlight

hockeypuckhulyhyperswitchimmichivatarjellyfinkeycloakkeygenkubokuttlibrechatmailmailcowmariadbmatomomatrixmauticmetabases3managerexternal-s3mission-controlmongodbmonicamysqlnetbirdnextcloud

nginxopenhandsopenobservepaperlesspeertubepenpotplanepmmpostgrespostizpulprediss3shlinksonarqubesupabasetraefikuptime-kumavaultwardenwazuhwhoamiwindmillwordpress+ any Docker app

browse the app catalogue

shc install nextcloud

nextcloud gitlab immich librechat

supported platforms

aws azure gcp hetzner digitalocean linode proxmox cloudstack openstack scaleway vultr exoscale alibaba

linux macos docker podman incus apple

if one person can't run it, it won't last
Hundreds of great apps, locked away
An app is not a container

It is a container plus a hostname, a certificate, a database, an identity provider, a backup schedule, and a restore that has actually been tested.

DNS + TLS

--runtime docker|podman|incus|apple
Connect two apps, never touch a credential

files hold references, values stay encrypted in the vault
Sign in once

someone revoked nextcloud when they left. the leftover eight still take the old password.

without · a login per app
NO KUBERNETES Skip the control plane. One person is enough.

upgrade the control planepatch etcdback up etcd againrenew the control-plane certs babysit cert-managerrenew the kubeconfiguntangle RBACdebug CoreDNS again bump the CRDschase a CrashLoopBackOffre-tune the HPAresize the PVCs pin the CNI versiondrain the node firstcordon the bad nodeupgrade kubelet everywhere fix the webhook timeoutre-apply the helm valuesreconcile helm driftmigrate the ingress class audit the pod securitytighten network policiesunstick a terminating namespace read the operator changelogmind the version skew retry the cluster upgraderebalance the autoscaler test the velero restorere-taint the nodes

Kubernetes solved a problem most people don't have

migrate from

caprover casaos cloudron coolify dockge dokploy easypanel kube portainer runtipi synology truenas umbrel unraid yunohost

continue at your own risk

wire apps together

nextcloud

gitlab

keycloak

postgres

external-s3

mailcow

anatomy of an app

An app is the authored compose plus app.yaml. A deployment is that app installed for one tenant and one environment: the running unit, and it owns its network. --stack only names it. The leader renders each replica's spec once, at install or update, and stores it. Nodes materialise the stored spec. Each vars value is a colored thread into its compose {{ }} marker, the same color on the value, the marker, and the resolved fill. Activate any pair; the others stay lit. Click empty space to reset.

app.yaml

name: gitlab
vars:
  rootPassword@secret!
  postgres:
    database@string!
  images:
    gitlab@string!

vars.yaml

rootPassword: 
postgres:
  database: 
images:
  gitlab:

compose.yaml

services:
  gitlab:
    image: ""
    environment:
      GITLAB_OMNIBUS_CONFIG: |
       db_database = ''
       root_password = ''
  valkey:
    image: "{{ vars.images.valkey }}"

shc install gitlab

An app is the authored compose plus app.yaml. A deployment is that app installed for one tenant and one environment. It owns its network. --stack names it. The leader renders each replica once, at install or update, and stores the spec. Nodes materialise the stored spec. apps/gitlab/{app,vars,compose}.yaml

highly available

rqlite3 voters · healthy

select a node to operate on it

3 / 5

survives node failure

reconcile pass · 10s here · 1m in prod next in 10s

deploymentnextcloudrunning

servicenextcloudrunning 3/3

replicanextcloud-1running

replicanextcloud-2running

replicanextcloud-3running

two ways in

your data, your choice

backendsdisk on the nodenfs · cephfshcloud · awsebs · gcepd · +6 more

node-2app node

on-node
rwo

shared FS rwx

single
attach

node-5standby

nextcloudhealthy vol-0abc123

attached · node-2 · rwo

node-2 disk · the volume is part of the node; it can't move, and node loss loses it shared FS · every node mounts the same bytes: nothing to move, ever block device · drag it to node-5: it detaches, reattaches, and the app follows

shc update nextcloud --where 'node=node-5' --yes

shc update nextcloud --where 'node=node-5' --restore-from-backup --yes

The bytes sit on that node's disk. Lose the node and you restore from backup. Every node mounts the same filesystem (RWX), so a move is never needed, and never possible. No auto-failover by default (auto_failover: false): a block volume reattaches only after you move the app.

Drag the volume onto a node to attach it there, or use the left and right arrow keys to aim and Enter to drop.

Drag nextcloud onto a node to run it there, or use the left and right arrow keys to aim and Enter to drop.

backups without downtime

A backup you haven't restored is a hope. Snapshots that were never replayed are the most common way a self-hosted estate dies. Backups here are scheduled, encrypted, and restorable with one command.

frozen copyconverter · pg_dumppostgres.sql

one backup holdsvolumespostgres.sql

Without converters a deployment snapshots its volumes and nothing else. shc backup export writes a backup out to a local archive.

shc backup create -s nextcloud

the runbook

Five commands stand a three-node cluster up and land nextcloud and postgres on node-1, wired to each other. The first node prints a token with shc node token; the next runs shc node join. Placement is --where. A published route is --expose (--host is a hidden gone flag). --scale is <name>=<n>.

leadernode-1

nextcloud

postgres

voternode-2

voternode-3

3 voters · quorum 2

shc init

shc node join <token>

shc install postgres

shc install nextcloud -i postgres

3 nodes · 2 apps · healthy

boot sound “CRT turn on” by Coolshows101sound · CC BY 4.0 · trimmed

Selfhosted Cloud▋

supported platforms▋

Hundreds of great apps, locked away

An app is not a container

Connect two apps, never touch a credential

Sign in once

Kubernetes solved a problem most people don't have

migrate from▋

wire apps together▋

anatomy of an app▋

highly available▋

survives node failure▋

two ways in▋

your data, your choice▋

backups without downtime▋

the runbook▋