For a new home self-hosting setup, choose k3s: it is Homestead's preferred and most widely tested platform.

The cluster guides - k3s, RKE2, Harvester, an existing cluster - each end with Homestead running. This page is what they share: what got installed, the first sign-in, and looking after Homestead afterwards.

Everything Homestead creates is named homestead-* or carries homestead.io/* labels and annotations, so it is easy to find with kubectl. There is one Homestead per cluster. The plain manifest installs it into lab; the Helm chart into its own namespace, with your apps in lab.

Both install the same objects - the chart is generated from the manifest.

• Helm (oci://ghcr.io/homestead-lab/charts/homestead) takes values instead of edits, and helm uninstall removes it (your apps, their namespace and Homestead's data volume are kept).
• The manifest (deploy/deploy.yaml) needs only kubectl, and is what the k3s script uses.

Updates are the same either way: Homestead updates itself, below.

The interactive installer enables a Host Console by default on k3s and RKE2 servers and workers. It replaces the first local console's login screen with live CPU, RAM, uptime, IP addresses and local disk usage. Server nodes also show cluster members, readiness, Kubernetes versions and Homestead's service address.

The dashboard uses green for healthy nodes, amber for warnings or cordoned nodes, and red for unavailable APIs, failed nodes or high resource usage. CPU, per-core load, memory and disk usage have coloured bars, with a rolling CPU history graph. Usage turns amber at 75% and red at 90%. Narrow terminals stack the panels and keep the full view available by scrolling.

The logo, history graph and usage bars use Unicode Braille dots, like btop. The house outline and two white infrastructure layers are rasterised from Homestead's logo, sized for tall terminal cells. Press A to switch between dots and plain characters if the console font lacks Braille glyphs. Non-Unicode locales use plain characters automatically; host-console.py --ascii also forces that view. Workers show local health; cluster details require a server's local kubeconfig. Harvester keeps its native console.

Press Enter, Q or Escape for the normal authenticated login prompt. Use the arrow keys or Page Up/Down to scroll. The screen returns after logout. SSH and other consoles work as usual. Host metrics refresh every two seconds; cluster status refreshes every fifteen seconds and shows API failures.

Choose No at the Host Console question, or set HS_CONSOLE=no for unattended installation. To enable or disable it later, use Configure host status console in the installer menu, or run the downloaded installer:

sudo sh install.sh --console
sudo sh install.sh --no-console

In Homestead, the host console is an add-on for the whole cluster: Settings → Cluster → Add-ons → Host console, with one On/Off switch. While it's on, Homestead keeps every Ready host on the console bundled with the running Homestead:

• it installs the console on hosts that don't have it, including hosts that join later
• it updates hosts when Homestead carries a newer console, so updating Homestead rolls the console out across the cluster

Turned off, it's removed from every host. Homestead works through a couple of hosts every ten minutes, and the row lists each host's state. A host that fails is tried again an hour later. On a cluster set up before this setting existed, the console counts as on where the installer turned it on. Harvester hosts keep their native console.

Changes take effect on the next boot or logout, so installation does not interrupt a logged-in administrator. The console needs systemd, agetty and Python 3 with curses; the installer installs Python if needed. It reads local metrics and Kubernetes status without displaying credentials or providing an unauthenticated shell. The screen is visible to anyone with physical or remote console access.

For manual removal, delete only /etc/systemd/system/[email protected]/50-homestead-console.conf, then run sudo systemctl daemon-reload. Normal login returns on the next boot or logout.

Open http://<address>:8088. The first visit asks for an administrator's name and password, and that account can then add others under Settings → Users and access → Add user:

Sessions last 12 hours of inactivity, or 30 days with Keep me signed in, and never more than 90 days. Changing a password signs that account out everywhere.

The first time an administrator signs in, Homestead opens its setup guide (/setup): six chapters - Is it healthy?, Reach it, Keep data safe, Make it yours, Connect and First apps - each step a sentence on why it matters, the cluster as it is now (a diagram where it helps: quorum, the VIP, copies across nodes, the HTTPS path from a phone), and one thing to do.

• Homestead checks each step itself, every time: a VIP set, backups configured, disks in use, the default storage class matching the nodes, notifications on. A step done and since undone shows again. Skip puts one aside; Bring it back returns it.
• Admins see every chapter. Everyone else sees the steps that are theirs: appearance, installing it on their phone, and notifications.
• It covers what new clusters most often miss: two servers (no node can fail until there is a third), getting HTTPS - needed to install Homestead on a phone and get notifications - with a Cloudflare Tunnel or Tailscale set up from the step, disks Longhorn is not using yet, a default class with too many or too few copies, an encrypted export of Homestead's own settings, and a second administrator.
• Until it is done, the Dashboard says Setup: 9 of 21; Hide from the Dashboard removes that, and Settings → Homestead → Setup guide always opens it.

The setup guide walks through all of these; in short:

1. Settings → Homestead - check Homestead's own health: every background task, the node probe, Samba, backup storage. See Settings.
2. Check node telemetry. Homestead installs the node probe automatically, unless explicitly declined in the installer. It adds temperatures, host devices, disks and SMART drive health. Review its status in Settings → Homestead.
3. Networking → Workload VIPs - keep a few addresses for apps (Harvester, or MetalLB). See Networking.
4. Data protection → Plans - snapshots and backups in one go. See Data protection.
5. Name it - Settings → Homestead → About this installation → Site name shows under the logo, handy with more than one cluster.

When a release comes out, a Homestead button with the new version appears on the top bar. It, and Settings → Updates, show the version you run, the new one with a link to its release notes, and the helpers Homestead runs beside it (the SMB and NFS servers and the object store used for moves). Update opens the same reviewed rollout apps get: helpers first, Homestead last, the old image kept for rollback. The page reconnects by itself while Homestead replaces itself. Homestead and its helpers are platform containers, so they are hidden on Containers and not counted among app updates; see Settings.

Each release carries the permissions it needs, and Homestead brings its own role up to date when it starts. An install older than that feature needs to be given that right once:

kubectl apply -f https://raw.githubusercontent.com/homestead-lab/homestead/main/deploy/rbac.yaml

Settings shows this line whenever Homestead finds it cannot update its role. Use the version-pinned command shown there. The RBAC-only manifest does not replace your Deployment, Service address, configuration or volumes. Avoid reapplying the full install manifest over a customised installation just to gain a new permission.

For a Helm-managed installation, update the existing release using its actual release name and namespace (shown by helm list -A). For example:

helm upgrade homestead oci://ghcr.io/homestead-lab/charts/homestead -n homestead --version 2.8.167 --reuse-values --set-string image.tag=2.8.167 --wait --timeout 5m

This preserves saved values while explicitly updating the image even if an older tag was pinned. If a HelmChart controller or GitOps manages the release, update its desired version/values there instead of competing with the controller.

After updating, check the installed version, Permissions, and Homestead's ready status. From v2.8.167, host power control defaults on but still requires an administrator's impact review and passes no safety blocker automatically. An explicit ENABLE_NODE_POWER=false stays disabled; changing an environment setting restarts Homestead, not the host.

The Service is plain HTTP, for your LAN. To reach Homestead from elsewhere - and to install it as an app with push notifications on your phone - put it behind HTTPS. The reference's Cloudflare Tunnel section covers doing that safely: Homestead can change anything in the cluster, so it belongs behind Cloudflare Access (or a VPN), never published bare.

helm uninstall homestead -n homestead

Your apps keep running, and so do the lab namespace and Homestead's data volume. Delete the homestead-auth Secret yourself if you want the accounts gone too.

Do not kubectl delete -f deploy.yaml to remove a manifest install: the manifest also declares the lab namespace, and deleting a namespace deletes every app and volume in it. Delete Homestead's own objects instead:

kubectl -n lab delete deployment/homestead service/homestead serviceaccount/homestead
kubectl delete clusterrolebinding/homestead clusterrole/homestead

The homestead-data claim is kept until you delete it from Volumes.

Cluster-wide things Homestead may have made stay too: the PriorityClasses of the priority tiers, and the hot-spare taint on standby hosts. Set every critical and low app back to Normal first (or remove priorityClassName from them yourself): pods running now carry on, but Kubernetes refuses every new pod that names a class that is gone, so an app still naming one cannot restart, update or move to another host. Then remove both; the second command takes the taint off every host that has it, and says not found for each one that does not, which is expected:

kubectl delete priorityclass homestead-critical homestead-low --ignore-not-found
kubectl taint nodes --all homestead.io/standby-