In Web 4 Is About Building Your Own Systems I said the clearest place to see Web 4 is a home. This is the longer version. How it actually works, with real parts, and why the result outlives any smart home company.

Start with the annoying stuff

Every makeover starts the same way. Walk through the house and count.

Count the remotes. Count the apps. Count the hubs blinking in a closet. Count the subscriptions that exist only so a light bulb can talk to a server in another country and back.

In a lot of houses the answer is a drawer full of remotes, a phone folder full of apps that each control one thing, and at least one device that already stopped getting updates.

None of those things know about each other. The AC doesn't know the balcony door is open. The lights don't know anyone is home. That's the problem worth fixing. Not "make it smart." Make it one system.

Connect everything

The workhorse is the ESP32. A tiny chip with Wi-Fi built in, a few dollars each (roughly $5 to $10, prices move). Flash it with ESPHome, which is open source, and describe what it should do in a short config file. No app, no account.

Put a sensor on an ESP32 and it reports temperature, humidity, light, whether a door is open. Put a relay on it and it switches things. Put an IR LED on it and it becomes a remote control.

That last one is how the AC gets in. Most minisplits are run by an infrared remote, which is basically a flashlight that blinks in a pattern. An ESP32 with an IR LED pointed at the unit can send the same blinks, and now the AC is a device on the network like everything else.

The rest of the house fills in around that:

• Sensors. Zigbee door and window sensors, temperature and humidity sensors. Small, battery powered, roughly $10 to $20 each.
• Presence. mmWave sensors, which can tell someone is sitting still on the couch. Old motion sensors think you left the moment you stop moving. A DIY one on an ESP32 runs maybe $10, an off-the-shelf one closer to $80.
• Lights. Smart relays behind the existing switches, so the wall switch still works and nobody has to learn anything. Roughly $15 to $20 per switch.
• Locks. The idea is a lock that speaks Zigbee or Z-Wave and works locally, not one that only answers to its maker's cloud.
• Power. Smart plugs that measure watts, so the washer can announce it's done.

The rule for every part: it has to work locally and speak an open protocol. If it only works through someone's cloud, it doesn't go in.

Add a home server

All of those devices need somewhere to report to. That's the home server.

It doesn't need to be big. A small mini PC on the network, running open source home automation software, hears from every sensor and decides what happens next. Nothing leaves the house unless you want it to.

Here's a test you can run on any smart home: unplug the internet. In a cloud smart home, the lights stop being smart. With a local server, nothing should change. The door sensor still talks to the server, the server still talks to the AC, the AC still turns off.

A private iPhone app

Off-the-shelf dashboards are fine. But this is the personal software era, so the idea is that the house gets its own app.

A private iPhone app that controls everything in that specific house. It has the rooms that house has, the buttons the people who live there actually press, and nothing else. No ads. No account. No subscription. It never has to be in an app store. It doesn't need to be.

This is the same idea as the menu bar app and the finance apps from the Web 4 post. An audience of one. Or for a house, an audience of however many people live there.

Scenes

Connecting everything is the plumbing. Scenes are where it starts to feel like something.

A scene is a set of things that happen together because of one trigger. A few you can build:

• Leaving. Last person out, presence goes empty, the house turns off lights and AC. No "did I leave the AC on" from the beach.
• Door open. Balcony door open for more than a minute, AC pauses. Close it, AC resumes. Small rule, and it ends an argument that has been going on since air conditioning was invented.
• Evening. Lights warm up and dim as it gets late. Same idea as the screen warmer on my Mac, applied to a room.
• Guests. If a house doubles as a rental, a check-in scene and a checkout scene can reset everything to normal between stays.
• Laundry. The washer's plug sees the power drop and sends a notification.

None of these need a new device. They're just rules over things that are already connected. That's the payoff of building a system instead of buying gadgets. Every new sensor makes every scene smarter.

No lock-in, no end of life

This is the part I care about most.

Smart home companies come and go. When they go, or when they decide a product line isn't worth supporting, the devices stop working or stop getting fixes. You did nothing wrong. Your hardware just became e-waste on someone else's schedule.

An open source house doesn't have that problem. ESPHome configs are text files. The server runs open software. The protocols are open. If a sensor dies, any sensor that speaks the same protocol replaces it. If a better chip comes out, flash it and swap it in. There is no vendor to ask for permission and no company whose bankruptcy shuts off the lights.

It also upgrades in small pieces. Add one room this month. Add presence next month. Rewrite a scene in five minutes when you notice it's annoying. The house keeps getting better instead of slowly expiring.

Every house is different

I've started doing these makeovers for other people's houses, and filming the before and after. The thing I keep noticing is that no two systems end up the same. Different layouts, different habits, different things that drive people crazy. The parts are the same. The system is built around the people.

That's Web 4 again. Not a platform you fit inside. A system that fits around you.

If you want to try it, start with one annoying thing. The AC that runs with the door open. The light nobody turns off. Put one ESP32 on it. Then add a second thing. Then make them talk.