I run my own machine for artificial intelligence tasks at home, so the data stays inside the house instead of going to some provider.

This time I am not explaining anything. I am asking: how would you set such a server up, and what would you do differently than I did?

Suggestions from people who already run one are the most useful ones here.

Most of what we call a smart home today is still a set of rules somebody typed in by hand. That is about to shift.

In this clip I sketch where I think things are heading once systems can interpret a situation instead of just matching a condition.

Whether that feels helpful or slightly unsettling probably depends on the person.

I sat down with another smart home hobbyist to compare how each of us combines two tools: Home Assistant, the open hub that controls the house, and n8n, which chains services and steps together.

The pairing takes over where a plain automation runs out of room, and building a workflow is quick once you get the hang of it.

Different setups, similar conclusion. Neither of our houses works the way it did before.

A standard setup switches lights and reads sensors. With a bit of extra work it can do considerably more than that.

In this clip I go through the additions that made the biggest difference in my own house, the ones that turned a collection of devices into something that feels coordinated.

None of it requires starting over. Each piece can be added to what is already running.

When the panels on the roof produce more than the house consumes, the surplus goes to the grid. It is worth more to me if I use it right away.

So I built a dashboard that tracks production and consumption live, driven by an ESP32, a small cheap microcontroller, together with ESPHome and Home Assistant, the software running my house.

The result is deliberately simple: a traffic light that tells me whether now is a good moment to run the washing machine or dryer, or whether it is better to wait.

Reading a number is optional. Glancing at a colour takes a second.

Instead of looking at how much water is used in total, I look at how quickly it flows. That single change makes patterns visible.

A shower shows up as a clear peak. Something that keeps running when nobody is home shows up too, and that is the case worth catching early.

Simple analysis, useful result, and no extra plumbing involved.

Searching the house for a misplaced laptop is a waste of an afternoon, so I let the cameras do the looking.

Snapshots go to Ollama, a tool that runs image analysis locally rather than in someone else’s cloud. It checks whether the object appears in the picture.

What comes back is a plain sentence about where the thing is, for example that the laptop is on the table.

The same principle works for other items you routinely put down and then cannot find.

My Ray-Ban Meta glasses can send messages, so I gave them something useful to talk to: a chatbot on WhatsApp that sits between me and my home.

The bot passes what I say along to Home Assistant, the open-source software that runs all my devices. Speaking a request out loud is enough to trigger it.

The nice part is that nothing new had to be worn or carried. The glasses were already on my face, and the messaging app was already there.

Some gadgets exist purely because someone could build them. This one opens bottles and connects to a home automation system over Matter, the shared standard that lets devices from different brands work together.

So yes, a beer can now be opened at the press of a button, and the rest of the house knows about it.

Whether anyone actually needs that is a separate question. I found it absurd enough to be worth showing.

Magnetic child locks are safe, but hunting for the little magnet every single time got old fast. So I automated the locks instead.

Now the cupboard opens from my smartwatch or phone, or by holding an NFC tag against it. NFC is the same short-range contact technology used for card payments.

It keeps the kids out of the drawers they shouldn’t reach, and it doubles as one more small security layer in the house.

Not every smart home project has to be useful. This one is mostly about how the room looks.

I built an LED matrix, a grid of individually addressable lights, and drive it with WLED, free software for controlling LED strips and panels.

It sits on the wall and does its thing, and the effect is far nicer than I expected from a handful of small lights.

This project sat on my list for years. Every time I saw someone else’s glowing wall panel I thought about it again, and then put it off.

It is finally built and hanging on the wall. Seeing the thing actually running, instead of picturing how it might look, turned out to be worth the long delay.

This is a short look at the finished result, nothing more than that.

The automations I like best are rarely the complicated ones. They are the small routines that quietly remove a bit of friction every day.

This is one of my favourites, and I walk through what it does and why it is set up the way it is.

No special hardware, no clever trick. Just a rule that has kept working without me thinking about it.

People often assume a smart home needs elaborate logic before it pays off. My experience is the opposite.

I built this one in Home Assistant, the open-source platform I use to tie my devices together, and it is about as plain as an automation gets.

It just makes an everyday moment slightly easier, and that is the whole point. Simple things are the ones that survive.

Phone screens keep growing, and the top row of icons has drifted out of thumb range for anyone holding the device in one hand.

iPhones have a built-in answer for that: a gesture that slides the whole screen downwards so the upper half comes to you.

It is not a secret feature, but plenty of people have never come across it. From an interface design point of view it is a neat piece of problem solving.

The self-built remote I have been working on took another step forward: it can now send infrared signals as well.

That matters because plenty of older devices in a living room still only listen to the same invisible light pulses a classic remote uses.

So one handmade device covers both the modern connected gear and the equipment that never became smart in the first place.

Weeks go by and the small everyday moments blur together. I wanted a record of them without having to sit down and write one.

So I have generative AI, software that produces text on its own, pull together what happened in the family over the past week or month and turn it into a diary entry.

The result is a running log of ordinary family life that writes itself while we get on with things.

Instead of a fixed playlist, my home now produces fresh music of its own each day.

A generative AI creates the tracks, and what it makes depends on the situation: who is currently at home and which scene, meaning which preset mood or setting, is active at that moment.

The soundtrack is therefore never quite the same twice, and it shifts with the room rather than looping in the background.

Deciding what to cook is one of those daily questions that eats more energy than it should.

So I handed it over. A generative AI, meaning software that comes up with content by itself, proposes the menu, and the suggestion is waiting when we get home.

It does not cook anything, but it takes the blank-page moment out of dinner planning. Ignoring the suggestion is always an option.

Once enough automations are running, it gets hard to say what the house actually did all day.

I now have a generative AI write a short recap of the previous day, so all those background routines become readable in one place.

The useful part is what stands out: anything unusual or out of pattern shows up in the summary instead of staying buried in the logs.