The short answer
Choose on what you must install on the host, and which inbound ports you can actually give it — not on a performance claim. We have not benchmarked any of the three, and the minimum-RAM numbers published by third-party comparison sites disagree so widely that quoting one would be guesswork dressed as a specification.
What you can verify before installing is more useful anyway:
- Home Assistant is the only one of the three that offers a complete operating system image as its recommended installation type. Its own documentation lists HA OS and HA Container as two distinct types, and says Container installations cannot use Apps — which is where Thread and Z-Wave support lives. Its web interface is reached on 8123.
- openHAB is written entirely in Java and depends on a 64-bit Java 21 JVM that you install on the host. Its own
openhab-cli infoexample prints 8080 for HTTP and 8443 for HTTPS. - Domoticz is a C++ binary with an HTML5 web frontend, documented on 8080 for HTTP and 443 for HTTPS, and it ships with a default admin / domoticz login that its own wiki tells you to change immediately.
That last pair is the reason this page exists. openHAB and Domoticz both want 8080. Domoticz also wants 443, which is almost certainly the port your reverse proxy already owns. On a mini PC that already runs Docker, a media server and an edge proxy, those collisions are the first thing that will actually bite you — long before any difference in memory footprint matters.
What we measured, and what we did not
This is a documentation-and-port-planning comparison. We did not install Home Assistant, openHAB or Domoticz on the Beelink EQi12 units we own. We did not measure their boot time, memory footprint, CPU usage, database growth, automation latency or device capacity. Under our own evidence rules that makes all three untested applications, so this page contains no performance numbers for them and no sentence of the form “we tested the hub and found…”.
Where a number belongs to a project rather than to us, it is labelled as a documented requirement or a vendor statement, with the source and the date we checked it. Domoticz’s own site says it uses under 50 MB of RAM and runs on a Raspberry Pi Zero; its manual says 256 MB of memory is recommended. Those are the vendor’s claims about its own software, not an independent benchmark and not something we verified.
Where a number belongs to us, it describes the EQi12 host, not the hub: wall power, network throughput, container memory and a long stability run. Host headroom is a real measurement of a real machine. It is not evidence that any particular smart home application will use a particular amount of anything. We keep those two categories separate throughout, because conflating them is how a spec sheet turns into a fake benchmark.
Everything in the official-facts table below was read from each project’s own documentation on 12 November 2026. Documentation changes; check the linked pages before you act on a port number.
Evidence 1 — documented installation types and runtimes, side by side
| Home Assistant | openHAB | Domoticz | |
|---|---|---|---|
| Official installation types | HA Operating System (recommended) and HA Container | Platform-independent archive, or package repository (apt/yum); openHABian image for Pi | Linux installer script, Windows installer, Docker image, or compile from source |
| Runtime the host must provide | Python application; OS image or Docker container | 64-bit Java 21 JVM explicitly required | C++ binary; no JVM required |
| Default HTTP port | 8123 | 8080 | 8080 |
| Default HTTPS port | Not a documented default on the pages checked | 8443 | 443 |
| Config model | UI-first, with YAML for advanced work | Files plus UI; pluggable bindings on an OSGi runtime | Web UI, plus dzVents, Lua, Python and Blockly scripting |
| Notable documented caveat | Container installs have no Apps; Thread and Z-Wave lack out-of-the-box support there | JVM version is a hard prerequisite, not a suggestion | Ships with admin / domoticz credentials; 2025.2 requires Debian 12 Bookworm with OpenSSL 3 |
| Source checked 2026-11-12 | installation page and Linux install guide | installation overview and openhab-cli info example | Getting started wiki, Raspberry Pi wiki and official manual |
Two details in that table are easy to miss and worth more than any benchmark.
The first is the HA OS / HA Container split. The official installation page names exactly two installation types and recommends the OS for most users. It then states that Container installations do not have access to Apps, and names Thread and Z-Wave as integrations controlled by Apps with no out-of-the-box support on Container. If your plan involves a Zigbee or Z-Wave coordinator and you assumed “Docker is Docker”, this line decides your layout before you flash anything. The Linux guide also uses --network=host for the container and points at http://<host>:8123, which is a different networking model from the port-mapping you may be used to.
The second is Java 21 as a host dependency. openHAB’s installation overview says it is written completely in Java and that you need an up-to-date 64-bit OS with a Java 21 JVM platform installed. It even recommends Eclipse Temurin or Azul Zulu if your distribution does not package one. On a Windows mini PC already running Docker Desktop, that is a second runtime to install, patch and keep compatible — a real maintenance cost that a “lightweight vs heavyweight” comparison completely ignores. Domoticz needs no JVM at all; that is a genuine structural difference, not a marketing claim.
Evidence 2 — the port map against a real Windows + Docker host
Here is what each project’s documented defaults look like when placed next to the ports a working mini PC already uses. The right-hand column comes from our own port audit and from the official default ports of the reverse proxies we documented earlier this week.
| Port | Who documents it | On our EQi12 host |
|---|---|---|
| 80 | HTTP entrypoint for reverse proxies; Caddy redirects it to 443 | Part of the 25 unique listening TCP ports / 36 bindings we recorded |
| 443 | HTTPS for a reverse proxy; also Domoticz HTTPS | Owned by the edge proxy in the usual layout — a direct conflict |
| 81 | Nginx Proxy Manager admin UI | Not a documented hub default |
| 8080 | openHAB HTTP and Domoticz HTTP; Traefik dashboard in a development-only insecure mode | A common admin port; a collision between the two hubs themselves |
| 8443 | openHAB HTTPS | Isolated from 80/443 |
| 2019 | Caddy admin API (localhost) | Loopback only |
| 8123 | Home Assistant web UI | Typically free |
The audit behind that column is a real Windows 11 plus Docker Desktop host on one of our EQi12 units: 25 unique listening TCP ports across 36 bindings, with 169 enabled inbound allow rules in the firewall. We published it because the binding scope — 0.0.0.0 versus 127.0.0.1 — turned out to matter more than the port number. A hub that binds 8080 on all interfaces will not coexist with another process binding 8080 on all interfaces, and a hub that wants 443 will not coexist with the reverse proxy that already terminates TLS for everything else.
This is also where the previous two comparison pages on this site pay off. Traefik vs Nginx Proxy Manager vs Caddy documents 80, 81, 443, 8080 and 2019 from each project’s own quickstart, and AdGuard Home vs Pi-hole covers the DNS-side port conflicts on the same class of host. Put those tables next to this one and a pattern appears: the hard part of self-hosting several services on one box is not CPU, it is that every project assumes it owns the machine.
Home Assistant’s 8123 is the quiet exception. It sits outside the 80/443 web pair and outside the 8080 admin cluster, so on an already-busy host it asks for fewer decisions before the first boot. That is a port observation, not a recommendation — but it is the kind of observation that saves an afternoon.
Evidence 3 — host headroom, and the arithmetic we will not overclaim
Our EQi12 is an Intel Core i3-1215U (6 cores, 8 threads) with 16 GB of DDR4-3200, a 512 GB WD PC SN540 NVMe drive, two Realtek gigabit ports and an Intel AX200 wireless card. These are measurements of that machine under a Windows plus Docker workload, published under CC BY 4.0 in the measurement data repository:
| Host measurement | Value | What it does and does not tell you |
|---|---|---|
| Docker idle wall draw | 12 W (14 W in the idle band we recorded at the wall) | Baseline power of our stack, not of a smart home hub |
| QSV transcode wall draw | 13 W | A media workload on the iGPU, not an automation workload |
| Container memory | 320 MiB across four healthy containers, against a 7.638 GiB container ceiling | Headroom on our stack; says nothing about a hub’s footprint |
| Wired throughput | 891–941 Mbps moving 4 GiB in both directions | LAN capacity, not hub traffic |
| NVMe sequential read | 623 MiB/s | Storage path, not a database write rate |
| Long stability run | 12.57 hours, zero HTTP failures, zero container restarts | Reliability of our stack over that window, not a hub uptime claim |
There is one arithmetic statement we will allow ourselves, and it is labelled as arithmetic: on that host, four containers using 320 MiB sit well inside a 7.638 GiB ceiling, so memory was not the binding constraint for the workload we ran. That is a statement about the workload we ran. It is not a prediction that any of these three hubs will fit in a particular amount of RAM, and we will not turn it into one. If someone quotes a minimum-RAM figure to you, ask which version, which installation type, which integrations and whose machine produced it. Our own earlier work on mini PC RAM budgeting explains why component-level numbers must not be extrapolated into an application result.
One more caveat worth carrying over: on the same host, tight docker stats polling inflated measured idle CPU by 29.7%. If you install a hub and then measure it with an aggressive monitoring loop, part of the number you see belongs to the observer. Our 24-hour stability samples and the 12 W power baseline are published with their sampling method attached for exactly this reason.
Where the three projects genuinely differ
Strip away the unverifiable performance claims and three real axes remain.
Installation surface. Home Assistant can be the whole operating system, which is a different commitment from installing an application. openHAB is a Java distribution that sits on an OS you manage and needs a JVM you keep current. Domoticz is a small native binary with a web UI and an installer script. These are three different answers to “what am I actually maintaining for the next three years” — a question that matters more on a mini PC that is already your file server, media server and DNS sinkhole.
Radio and integration model. Home Assistant’s own documentation ties Thread and Z-Wave support to Apps, which Container installations do not have. openHAB’s site describes a pluggable binding architecture and claims support for hundreds of technologies and thousands of devices — a vendor statement about its own ecosystem, not a count we verified. Domoticz’s wiki describes hardware gateways such as Zigbee2MQTT and Zwave-JS-UI that you add separately, and its site claims 150+ supported devices. If your device list is fixed and short, all three can plausibly serve it; if you are buying hardware next year, the ecosystem question is a bet on future maintenance, so read each project’s current integration list rather than a 2024 blog post.
Migration cost. None of the three shares a configuration format with the others. Moving later means re-pairing radio devices and rebuilding automations, because a coordinator’s pairing lives in the platform’s own database. That is not something we measured, but it follows from how gateways and device databases work, and the Domoticz wiki confirms that devices are imported and named inside its own UI. It is a strong argument for reading the installation-type caveats before the first install, not after you have twenty automations.
Decision table
| If your situation is… | Read the documentation for… | Because… |
|---|---|---|
| You want the least host setup and plan to use Add-ons, Thread or Z-Wave | Home Assistant OS | Official recommended type; Apps are included, unlike Container |
| You already run Docker and want no extra OS image | Home Assistant Container, with the Add-on caveat checked | Container has no Apps; Thread / Z-Wave not supported out of the box |
| You want a mature rules engine and are comfortable maintaining a JVM | openHAB | Java 21 is a documented host prerequisite |
| You want a native binary with no JVM and modest hardware | Domoticz | C++ binary; vendor states a very small footprint |
| Your reverse proxy already owns 443 | Check Domoticz HTTPS before choosing it | Domoticz documents HTTPS on 443 |
| You plan to run two hubs, or a hub plus a Traefik dashboard | Re-plan 8080 first | openHAB and Domoticz both default to 8080 |
Deployment checklist on an already-busy host
- Inventory your listeners before installing anything. Our Windows home server port audit and the port audit checklist generator exist for this. Record the port, the process and the bind address, not just the port number.
- Decide who owns 443. If an edge proxy terminates TLS, a hub that documents HTTPS on 443 needs either a different port or to sit behind the proxy. Traefik vs NPM vs Caddy maps the 80/443/81/8080/2019 defaults.
- Resolve the 8080 clash explicitly. Either pick one of openHAB / Domoticz, or change one of their ports and write down where you changed it. Do not discover this at first boot.
- Check the runtime prerequisite. For openHAB, confirm a 64-bit Java 21 JVM is present with
java -version. For Home Assistant Container, confirm you accept host networking and the no-Apps limitation. - Change every shipped credential. Domoticz’s wiki names its default login and tells you to replace it. Do it before you forward any port or expose the UI to the LAN.
- Plan the backup path first. A hub’s device database and automations are the part you cannot recreate quickly. Our Docker Compose backup and restore drill shows a restore test that caught a bug which faked data loss, and the backup retention planner sizes the window.
- Budget the always-on draw. Whatever the hub uses, it is added to a host that never sleeps. The home server power cost guide and power cost calculator turn our measured 12 W idle baseline into an annual figure with your own electricity rate.
- Measure after deployment, on your own host. If you need a footprint number, take it from your own monitoring with a sane polling interval — and remember the 29.7% observer effect.
The numbers and where they come from
| Statement | Origin | Category |
|---|---|---|
| HA OS recommended; Container has no Apps; Thread / Z-Wave not out-of-the-box on Container; Pi 4/5 minimum 2 GB RAM | Home Assistant official installation page, checked 2026-11-12 | Official documentation |
http://<host>:8123, --network=host, open TCP 8123 for the frontend | Home Assistant official Linux install guide, checked 2026-11-12 | Official documentation |
64-bit OS with Java 21 JVM; openhab-cli info prints 8080 / 8443 | openHAB official installation overview, checked 2026-11-12 | Official documentation |
| Default web UI 8080, HTTPS 443, default admin / domoticz, 2025.2 needs Debian 12 + OpenSSL 3 | Domoticz official Getting started and Raspberry Pi wiki, checked 2026-11-12 | Official documentation |
| Under 50 MB RAM, Pi Zero capable, 150+ devices, dzVents / Lua / Python / Blockly | Domoticz project site | Vendor statement, not independently verified |
| 25 unique listening TCP ports, 36 bindings, 169 enabled inbound allow rules | Our Windows port audit | A — directly measured on EQi12 |
| 12 W Docker idle, 13 W QSV transcode, 320 MiB / 7.638 GiB, 891–941 Mbps, 623 MiB/s, 12.57 h zero failures | Our measurement data repository | A — directly measured on EQi12 |
| RAM headroom arithmetic | Derived from the measured container figures | Derived arithmetic, not a measurement |
| Hub performance, footprint, latency, uptime | Not measured | C — deliberately left blank |
Where to go next
If you are still deciding whether a mini PC is the right box at all, the Raspberry Pi 5 vs mini PC comparison has our measured watts and throughput against 2026 Pi pricing, and Windows + Docker vs Proxmox covers the case where a hub is better placed in a VM. For the hardware itself, the EQi12 review is the canonical spec sheet we hold every other page to.
For sizing work, the Home Assistant hardware capacity estimator lets you check a box against published minimums and estimate recorder and camera storage with editable assumptions; the self-hosting alternative selector places this decision next to the other services a mini PC usually runs. If you are new to running several containers, the Docker Compose home server stack walks through a working stack with health checks and persistent volumes, and the 135-measurement lab index links every number we do claim back to its raw record.
The short version: pick the hub whose installation surface and runtime you are willing to maintain, give it ports that do not collide with the services you already run, change its default credentials, and back up its database before you add the first device. Then measure it on your own machine.