What You Can Actually Do with AI in Home Assistant Today

AI in Home Assistant isn't just hype anymore. Since the 2025.8 release, Home Assistant has real, usable AI features that go beyond novelty. But most people either don't know they exist or aren't sure what's actually practical vs. what's still a gimmick. Here's what you can do right now.

AI Tasks: The Foundation

AI Tasks landed in Home Assistant 2025.8 and it's the most useful automation building block for AI right now. Think of it as a way to hand a small job to an AI model and get a structured result back. You can use it inside automations, scripts, and templates.

The key thing: it's opt-in. You choose whether to use a cloud model (OpenAI, Google, Anthropic) or run one locally with Ollama. Home Assistant doesn't force AI on anyone, and your basic automations keep working without it.

What People Are Actually Using It For

Smarter notifications. Instead of getting a dry "Motion detected at front door" message, AI can look at a camera snapshot and tell you "There's a delivery driver at your front door holding a large package." People are using this for security cameras, pet monitoring, and even checking if the garage door was left open.

Camera analysis. This is one of the most popular uses. Point a camera at your driveway and ask AI if there's a free parking spot. One person set up a camera on their chicken coop to count the chickens. Another uses it to check if their cat is inside or outside. It's surprisingly practical once you start thinking about what your cameras can already see.

Fun notifications. Some people are using AI to generate creative or funny laundry-done messages, weather summaries in the style of a pirate, or bedtime reminders for the kids written as a short story. It's not essential, but it makes the smart home feel more personal.

Automation summaries. If you've built dozens of automations and lost track of what does what, AI can read through them and generate plain-language descriptions. Handy when you come back to a setup after months and can't remember why something exists.

Voice Assistant + AI

Home Assistant has been building its voice assistant (Assist) since before the AI wave hit, and it works well for direct commands like "turn off the kitchen lights" or "set a timer for 10 minutes." What AI adds is the ability to handle fuzzier requests.

Say "it's dark in here." Assist on its own doesn't know what to do with that. But with an AI agent behind it, it can figure out you probably want the lights on and act accordingly.

If you configure an LLM conversation agent and enable "Prefer handling commands locally," Assist tries the built-in intent matching first (fast, local, no AI needed). If it can't handle the command, the request gets passed to the AI agent. This means simple commands stay snappy and only the tricky ones hit the AI model.

Dual Wake Words

Since Home Assistant 2025.10, you can set up two wake words on each voice satellite. "Okay Nabu" runs one pipeline, "Hey Jarvis" runs another. This is great for multilingual households, but also useful if you want one wake word for quick local commands and another for AI-powered conversations.

AI Personalities

You can give your voice assistant a personality by writing a custom prompt for the AI conversation agent. Tell it to respond like a butler, a surfer, or a grumpy robot. It still handles the same Home Assistant commands, it just adds character to the responses. Kids love this.

Local vs. Cloud: Your Choice

Running AI locally with Ollama on a decent machine (or even a Raspberry Pi 5 for smaller text models — image and vision tasks usually want more CPU/GPU headroom) means nothing leaves your network. The tradeoff is speed and capability. Cloud models like GPT-4 or Claude are generally more capable but cost money per request and send data out.

Most people start with a cloud model to see what's useful, then move the things they use daily to a local model. Home Assistant makes it easy to swap between them.

OpenRouter Integration

If you want to experiment with different AI models without setting up accounts everywhere, Home Assistant added an OpenRouter integration. It gives you access to over 400 models through a single API. Useful for testing which model works best for your specific use case before committing.

What's Not Ready Yet

Let's be honest: AI-powered automations that "learn your habits" and auto-adjust your home aren't quite there yet in a reliable, set-it-and-forget-it way. The building blocks are in place, and the community is experimenting, but if you're expecting your house to figure out your routine on its own, that's still coming. What works today is using AI as a tool inside automations you've designed yourself.

AI in Home Assistant is real, practical, and entirely optional. Start with one use case, like smarter camera notifications or a voice assistant personality, and see if it adds value. If it doesn't, turn it off. That's the beauty of it.

What's the most creative AI use you've seen in Home Assistant? Share it in the comments.