AI is no longer just something you use on your phone or laptop. It's quietly making its way into smart homes — and not in a gimmicky way. We're talking about a home that can explain itself, fix its own problems, suggest things you never thought to set up, and even see what's happening outside your front door. Here's a look at what's actually available right now, and why it's more impressive than most people realise.

Talking to Your Home Like You Mean It

Most smart home voice assistants work like a very rigid vending machine — say exactly the right phrase, get exactly the right result. Anything outside the expected format and you hit a wall.

That's changed. Home Assistant now supports AI-powered conversation — you can connect it to OpenAI, Google Gemini, Anthropic's Claude, or even a local model running on your own hardware. The difference is immediately obvious. Instead of memorising trigger phrases, you just say what you mean.

• "Turn off everything downstairs except the hallway light." — Done.
• "Is the back door still open?" — It checks and tells you.
• "Set the living room to movie mode." — It figures out what that means based on your devices.

No rigid command structure. No separate automation needed for every combination. You just talk, and it listens.

Claude as Your Home's Brain

Taking it a step further — what if your AI assistant could actually reach into your home, look around, and act on what it finds?

That's what's possible when you connect Claude to Home Assistant using a tool called ha-mcp. Think of it as giving Claude a live window into your home: your devices, your automations, your history, your dashboards, your logs. Not just a conversation — actual access to act on things.

Automations you didn't know you needed
Claude can look at everything connected to your home and spot the gaps. It might say: "You have a door sensor, a motion sensor in the hallway, and smart lights — but none of them are connected. Want me to create a routine that turns on the hall light when the door opens after dark?"
You say yes, then add "but only after 6pm, and off again after two minutes" — and it rewrites the automation on the spot. No editors. No documentation. Just conversation.

Tweaking what you already have
Existing automations are just as easy to adjust. "Make that one only run on weekdays." "Add a 15-minute delay." "Take the bedroom off that lighting scene." Claude reads the automation, adjusts it, and saves it back — you just describe what you want changed.

Troubleshooting why something stopped working
This is where it gets genuinely useful. Automations fail silently all the time — a condition that's no longer met, a device that went offline, a trigger that changed after an update. Normally, figuring out why means digging through logs and developer tools.
With Claude, you just ask: "My front door automation hasn't fired in three days — what's going on?" It pulls the execution traces, reads through the recent logs, and tells you in plain English what happened. Then it fixes it.

Dashboards that actually make sense to you
Claude can build and modify dashboards conversationally. "Give me a view for the bedroom — just the lights, the temperature, and the blinds." No manual layout editing required. Your dashboard becomes something you shaped with your words, not something you wrestled into existence.

The icing on the cake
Claude is already impressively capable on its own — but Home Assistant has its own preferences. Ways it likes automations structured. Built-in tools that work better than workarounds. Patterns that are easier to maintain long term.
There's an optional add-on called Home Assistant Agent Skills that teaches Claude exactly this. Think of it as giving Claude a Home Assistant expert sitting next to it. Same AI, but now it knows HA's opinions. The result is automations that don't just work — they work the way Home Assistant was designed to handle them. It's the difference between an AI that can do the job and one that's actually been trained for it.

We're actively running this in our own setup — it's early days, but the potential is hard to ignore.

Cameras That Actually Understand What They See

Standard motion sensors don't discriminate. A shadow, a cat, a leaf blowing past — all of it triggers the same alert. After the third false alarm, most people turn notifications off entirely.

AI changes this completely. Frigate is a local video processing system that uses an AI model to identify what triggered your camera: a person, a car, a dog, a parcel left on the doorstep. Suddenly your notifications mean something. "Person at the front gate" lands very differently to "Motion detected."

You can go further still. Send a camera snapshot to an AI model with vision capabilities and ask it to describe what it sees. "A person in a dark jacket is standing near the letterbox." "A delivery has been left by the front door." It's the kind of detail that used to require someone watching a monitor.

Notifications That Don't Sound Like a Robot

There's something jarring about a smart home that shouts raw sensor data at you. "binary_sensor.backyard_motion triggered at 23:42." Cool, thanks.

With an AI in the loop, that changes. A language model can take raw home events and turn them into natural sentences before they're sent or spoken. "Hey, there's movement out the back — it's nearly midnight, might be worth a check."

Spoken through a smart speaker, that's a genuinely useful notification. The same system can generate a morning briefing: combining today's weather, your calendar, and your home's current state into a few natural sentences read aloud when you wake up. And the voice quality has improved enormously — modern text-to-speech options are a long way from the robotic monotone of a few years ago.

The Wow Factor Is Real

We're still in the early days of what's possible. AI in a smart home isn't about replacing your automations — it's about filling the gaps they can't. The things you never thought to set up. The questions your home can now answer. The problems it can resolve before you even notice them.

If any of this sounds like something you'd want in your setup, get in touch and we can talk about what's realistic for your home.

Hardware Used

• Home Assistant hub — any compatible device (Raspberry Pi, Home Assistant Green, or similar) that runs your smart home
• IP Camera — required for Frigate's AI object detection features
• Google Coral TPU (optional) — a small AI accelerator chip that significantly speeds up Frigate's object detection