Workbench
The Home Assistant Hardware I Would Buy Again—and What I Would Skip
I have 26 Caséta controls, six WLED controllers in one kitchen, a Matter-connected Eufy vacuum, and a local model running on an RTX 5070 Ti. That sounds like a shopping list. It is really a list of places where buying more hardware either solved the problem—or would make it worse.
My answer is shorter than my equipment list
This is not a ranking of everything that works with Home Assistant. It is the purchase filter I can defend from the equipment in my own house.
| Job | My answer | The boundary |
|---|---|---|
| Permanent wall-controlled lighting | Buy Caséta again | The light remains obvious and usable without Home Assistant. |
| A loose bedside or movable lamp | Use one Matter bulb | I still need the exact bulb model before recommending the one in this house. |
| A custom LED installation | Build with ESP32/WLED again | The controller, power supply, strip, wire, fuse, and enclosure have to be treated as one system. |
| A robot vacuum | Choose it for cleaning first | The S2’s Matter controls are useful, but not a reason by themselves to buy a flagship vacuum. |
| A security camera or doorbell | Wait for an exact model | HomeBase compatibility and Home Assistant exposure vary by device. |
| Local AI for Home Assistant | Use a capable PC I already own | I would not justify a new RTX-class GPU with home control alone. |
The easiest yes: Caséta for the walls
The strongest purchase decision in this house is also the least futuristic. I want a visitor to press the wall control and get light. I want that to remain true while Home Assistant is restarting, the network is down, or an automation is wrong.
That is why permanent lighting here uses Lutron Caséta. The live registry contains nine PD-6WCL dimmers, eleven PD-8ANS switches, three PD-FSQN fan controls, three Pico remotes, and one L-BDG2 hub. The older controls still work, so I would not replace them for cosmetic reasons.
For a new room, I would start with the current paddle-style DVRF-6L Diva smart dimmer for a compatible dimmable lighting load, the DVRF-5NS Claro smart switch when I need a neutral-wire switch, and the PD-FSQN only for a compatible pull-chain fan that needs four-speed control. The L-BDG2 hub is the bridge into Home Assistant.
The model numbers are not interchangeable. Load type, neutral availability, lamp compatibility, and whether the circuit controls a fan all come before automation. I keep the exact ratings and the current-versus-installed distinction in my Caséta and Matter buying guide.
A Matter bulb is an exception, not a lighting plan
A small lamp in a kid’s room is where a Matter bulb makes sense to me. There is no useful in-wall dimmer to install, the lamp can remain powered, and color or tunable white can add something that a normal bulb cannot.
I would not scatter smart bulbs behind ordinary wall switches. Somebody will switch off the power, the bulb will disappear, and the house will need an instruction manual for turning on a light.
I am deliberately not naming a bulb to buy yet. I know the role, but I have not rechecked the exact model in this house. “Matter bulb” is not specific enough for a product recommendation: the transport can be Wi-Fi or Thread, exposed features vary, and a Thread logo is not proof of Matter support.
I would build the WLED lighting again—with a parts list next time
The kitchen has six separate LED runs because that made the wiring and service boundaries sane. Home Assistant groups them for normal on/off and brightness. For the named white and color modes, one helper sends the same WLED JSON state directly to all six controllers.
I would keep that architecture. One unreachable controller does not stop the other five, and a newer scene request can replace the one already running. It solved a real problem that the common light abstraction did not: the strips do not all expose white in the same way.
I would not publish a cart full of generic ESP32 boards and power supplies from memory. The useful buying guide has to include the exact controller, strip voltage, power-supply capacity, injection points, wire, fusing, and enclosure. Until I recover those part numbers, the six-controller article is a control design, not a complete shopping list.
I like the S2 integration. I would not buy the vacuum for it.
My current Eufy Omni S2 appears in Home Assistant as a local Matter vacuum. I can start a room, several mapped areas, or the normal downstairs route. An alert tells me when it reports an error or spends too long looking for the dock.
That is useful. It is not enough to justify flagship-vacuum money. Home Assistant does not replace the Eufy app for maps, station settings, obstacle details, or model-specific cleaning choices. The live battery percentage has also contradicted the full-charge state, so I leave it out of automations.
I would choose the vacuum on cleaning, obstacle handling, dock maintenance, consumable cost, and repair support first. Matter is the bonus that lets the chosen vacuum participate in the house. The exact controls and rough edges are in my S2 setup.
Lightning changed how I treat the Eufy label
The old Eufy S1 Pro vacuum and video doorbell were both taken out by lightning. The current security base is HomeBase 3, model T8030, but a shared logo does not turn the base, vacuum, doorbell, and outdoor lights into one product family with one reliable Home Assistant boundary.
I am not recommending a replacement doorbell until I have the exact replacement model and can verify what it does with HomeBase 3. I am applying the same rule to the permanent outdoor lights: I know what is installed and how I want exterior lighting to behave, but I need the model from the device or app before linking a product.
The RTX 5070 Ti is useful because it was already here
My local Qwen3 model is fast enough to feel natural on an RTX 5070 Ti: about two seconds for a warm response in the measured setup. It can control exposed Home Assistant devices and use a bounded local memory list. It can also claim that it changed the lights when it did nothing.
I would not buy this GPU solely to run a home assistant. The sensible path is to try the hardware already available, watch actual GPU placement and response time, and keep the model behind tested scripts and scenes. The detailed measurements and the false-success failure are in the local Ollama article.
The purchase test I use now
- What still works when Home Assistant is unavailable? Essential controls should have a simple answer.
- Am I buying the core job or the integration? A vacuum has to clean; a dimmer has to control its load; a light has to make sense at the wall.
- Do I know the exact model? A product-family name is not enough for compatibility or a responsible link.
- Can I explain the failure boundary? Power loss, boot time, one-way radio, cloud dependence, and bad telemetry each need different handling.
- Would one purchase remove complexity? If the new device adds another app, radio, bridge, and recovery path without making the room better, I skip it.
The house does not need matching logos. It needs equipment with clear jobs. Caséta owns the walls. Matter fills a few lamp-sized gaps. ESP32 and WLED handle the lighting I cannot buy. Home Assistant coordinates the pieces without pretending it can erase their limits.
Disclosure: I own and use the equipment described here. The external links on this page go to manufacturer documentation, not affiliate links. If product links later pay a commission, I will label that relationship on the page.