Home Assistant Without Regret: A Setup Guide
Avoid common pitfalls in Home Assistant deployment by choosing the right install method, securing backups, bridging MQTT brokers, cleaning stale automations, and configuring safe remote access.

Home Assistant tends to become the most important thing in your lab without anyone deciding it should. It runs the lights, the thermostat, and the garage door. When it breaks, the household notices before you do. Treat it like production from day one to avoid regret later.
The book opens this chapter by highlighting four ways to install Home Assistant, noting that they are not equal. Many users fall into the trap of choosing a method that lacks long-term support or flexibility, only to find themselves maintaining deprecated software or managing complex dependencies manually. This guide covers the correct installation paths, how to back up data that actually survives disasters, how to bridge MQTT brokers without creating loops, how to clean out stale automations, and how to reach your instance safely from outside.
The trap: choosing a deprecated install method
The first trap is picking an installation method that the Home Assistant project no longer supports for new releases. There are four ways to run Home Assistant, and they carry different levels of risk and maintenance overhead.
Home Assistant OS in a VM is the recommended approach, particularly on Proxmox. It provides full Supervisor access, add-ons, one-click backups, and updates. The Container (Docker) method is fine if you are comfortable running add-on equivalents yourself, such as Mosquitto, Z-Wave JS, and others, as separate containers.
However, Core (a Python virtual environment) and Supervised were deprecated by the Home Assistant project in 2025. They are no longer supported for new releases. If you are on a Core install, you must plan a migration. The caveats pile up quickly: there are no add-ons, every Home Assistant release may require a newer Python than your OS ships, dependency builds break during upgrades, and you are on your own for backups of everything outside the config folder.
The move is straightforward but requires care. Make a full backup from the Core install by navigating to Settings → System → Backups. Stand up a Home Assistant OS VM, and restore that backup during onboarding. Keep the old instance shut down, not deleted, for a couple of weeks to ensure stability before decommissioning it.
The trap: storing backups in the same place
A backup stored on the same VM as Home Assistant protects you from a bad update. It does not protect you from a dead disk, a ransomware infection, or a house fire. You want three copies: the live system, a local backup, and one off-site.
Recent Home Assistant versions have automatic backups built in. Go to Settings → System → Backups, choose a schedule, how many to keep, and where to put them. Locations include network storage, which you must add first under Settings → System → Storage → Add network storage as an SMB or NFS share on your NAS. You can also use Home Assistant Cloud, which stores an encrypted backup off-site for subscribers. Additionally, add-ons and integrations for Google Drive, OneDrive, and others appear as extra backup locations.
Watch out for the encryption key. Home Assistant encrypts automatic backups with an emergency key. Download that key and store it in your password manager. Without it, an off-site backup is a very securely locked brick.
The Proxmox side helps too. The nightly vzdump of the Home Assistant VM gives you a whole-machine restore point. But that copy usually lives on the same NAS, so it isn't off-site. Pick one copy to send off-site; rclone to cloud storage from the NAS works well. Check it monthly to ensure the data is intact and accessible.
The trap: creating MQTT bridge loops
MQTT is the glue for a lot of home automation, including Zigbee2MQTT, ESPHome alternatives, custom scripts, and integrations from other systems. When you have more than one broker, such as Home Assistant's Mosquitto add-on and a separate broker on a container that a script publishes to, bridge them instead of pointing everything at one fragile box.
A Mosquitto bridge config on the sending broker looks like this:
# /etc/mosquitto/conf.d/bridge.conf
connection ha-bridge
address 10.30.0.30:1883
remote_username bridge_user
remote_password <use-a-long-random-password>
cleansession false
restart_timeout 10 30
# Topic pattern, direction, QoS, local prefix, remote prefix
topic sensors/# out 1 "" ""
topic commands/# in 1 "" ""
Things that trip people up include direction and retained messages. Direction is from the bridge's point of view. out sends local messages to the remote broker; in pulls remote messages down; both does both. Using both with the same topics on two bridges creates a loop. Retained messages and discovery can also cause confusion. If you bridge Home Assistant's homeassistant/# discovery topics, retained config messages copy across. That is useful once, but confusing when you delete a device and it keeps coming back. Clear a retained topic with mosquitto_pub -t -r -n.
Authentication is critical. Never run a broker with allow_anonymous true on a network other devices can reach. You can watch the state of bridges using mosquitto_sub -v -t '$SYS/broker/connection/#', and prove data is flowing by running mosquitto_sub -v -t 'sensors/#' on the far side.
The trap: letting stale automations accumulate
Over a couple of years, automations accumulate. You get ones for devices you no longer own, duplicates you made while experimenting, and ones that reference entities that were renamed. They clutter the UI, log errors, and occasionally fire at the worst moment.
Settings → Repairs catches some of this. For the rest, paste these into Developer Tools → Template to find automations that have never triggered:
{# Automations that have never triggered #}
{{ states.automation
| selectattr('attributes.last_triggered', 'none')
| map(attribute='entity_id') | list }}
You can also find entities that are unavailable right now:
{# Entities that are unavailable right now #}
{{ states | selectattr('state', 'eq', 'unavailable')
| map(attribute='entity_id') | list }}
Cross-reference these lists. An automation that has never fired and mentions an unavailable entity is a deletion candidate. Before deleting, disable it for two weeks. If nobody notices, delete it. If your automations live in YAML, commit the folder to git first, so "delete" is never permanent.
The trap: exposing port 8123 directly
Do not port-forward 8123 to the internet. Use Tailscale (covered in the next chapter), Home Assistant Cloud, or a reverse proxy behind a tunnel with strong authentication. If you use a reverse proxy, you must add its address under trusted_proxies in configuration.yaml, or Home Assistant will reject the forwarded requests.
The configuration looks like this:
http:
use_x_forwarded_for: true
trusted_proxies:
- 10.30.0.40 # your reverse proxy
Checklist
- Running Home Assistant OS (VM) or Container, not Core or Supervised
- Automatic backups on a schedule, with retention
- One copy off-site; emergency key saved in your password manager
- MQTT brokers authenticated; bridges with explicit topic directions
- Stale automations disabled, then deleted after two quiet weeks
- No 8123 port-forward; remote access through Tailscale, Cloud or an authenticated proxy


