26 physical controls + a bridge
Eleven wall switches, nine dimmers, three fan controls, and three Pico remotes. Permanent lighting remains usable at the wall.
Problems from my actual house
I made a ceiling fan smarter. Then it stopped turning on.
The useful part of home automation starts when the individually good products do something stupid together. I use Home Assistant to find those seams, fix them, and document what survived.
The whole setupWhat I would buyLutron CasétaESP32 + WLEDEufyPower and cost
Start here
Why my Caséta switch made a Bond-controlled ceiling fan unreliable
I tried tracking the RF state with five automations. Then I replaced that with a timed power-on routine. Both approaches were solving the wrong problem. The receiver needed to stay powered.
This is the system behind the articles
These counts come from the Home Assistant configuration and device registry I actually run—not a sample dashboard.
Six controllers, one light
Six separate LED runs appear as one kitchen light, with direct JSON commands for color temperature that fought Home Assistant’s normal light entity.
The current vacuum replaced the S1
I added the Omni S2 through Matter and built room, zone, whole-floor, and rescue-alert controls around what it actually exposes.
What I would spend money on now
What is installed here and what I would buy today are not always the same answer. These pages keep that distinction explicit.
What I would buy again for Home Assistant
Caséta is the easy yes. Matter bulbs, WLED, Eufy, and local AI each have a narrower job.
I have 11 PD-8ANS switches. I would buy Claro now.
The installed switches still work, so replacing them would waste money. For a compatible new neutral-wire circuit, I would check the current Claro model first.
Do you need the Caséta hub?
If Caséta is going into Home Assistant, yes. The useful question is which hub, not whether to skip it.
Would I buy HomeBase 3 for Home Assistant?
Only for the Eufy features I actually want. It is not a general-purpose Home Assistant hub.
Start with something that broke
These are finished builds, including the bad assumption or failure that made each one worth documenting.
Six WLED controllers, one kitchen light
Why the normal light entity fought white and color-temperature control, and the direct command path I kept.
The Eufy Omni S2 through Matter
Room and zone cleaning, the controls Matter leaves behind, and the alert I added when the vacuum cannot get home.
Ollama talks to Home Assistant
Measured Qwen3 speed on an RTX 5070 Ti, the setup that appeared to work but did not, and the memory boundary I kept.
Three automations that failed differently
An unpowered RF receiver, a controller still booting, and a vacuum that needed to admit it was stuck.
Software that failed in useful ways
The house is the main lab, but it is not the only place I find bad assumptions. These are failure stories from software I actively publish.
I took an M5Stick from 2.3 hours to about 12
A 40 MHz clock broke Bluetooth, manual light sleep caused watchdog resets, and ESP-IDF power management finally kept the desk display asleep and reachable.
My Bluetooth approval button broke while idle
The bridge delivered only the clock on Windows, then treated a quiet queue as a dead connection.
I built a tmux dashboard, then removed half of it
The useful cockpit appeared when live agent state replaced permanent screens for things I rarely opened.
My DNS editor kept the keys in the browser
I removed TSIG secrets from browser storage, then fixed a split-DNS path that could send an internal edit to the external view.
MediaWiki MCP could read after its login expired
Public reads made the service look healthy while protected writes had quietly lost their identity.
Two computers kept fighting over two monitors
DisplayCue coordinates the handoff without pretending DDC/CI can replace a hardware KVM everywhere.
Numbers I would rather calculate than guess
Can a smart plug pay for itself?
Subtract the plug’s own power use from the load it avoids.
What does an always-on device cost?
Useful for hubs, routers, mini PCs, and everything else running all year.
Will efficient hardware repay the purchase?
Compare the complete purchase price with the electricity it can actually save.