OpenClaw for your House

There’s a moment that Andrej Karpathy described on the No Priors podcast that stopped a lot of people mid-scroll. The OpenAI co-founder and AI researcher had just told his OpenClaw agent to find his Sonos system. Three prompts later, music was playing in his study.

“I can’t believe I just typed ‘can you find my Sonos?’ and that suddenly is playing music,” he said. “That’s crazy. That’s like three prompts.”

What he was describing wasn’t a product demo or a carefully staged prototype. It was what happens when you give a persistent AI agent access to your home network and let it figure things out. The agent scanned his local IP range, found the Sonos endpoints, reverse-engineered the API, and asked if he wanted to try it. Karpathy said yes. The music came on.

This is the moment the smart home industry has been circling for a decade without landing: a home that responds to intent, not just commands.

The App Drawer Problem

Most smart homes in 2026 are not actually smart. They’re complicated. There’s an app for your lights, a different one for your thermostat, another for your shades, a fourth for your security cameras, and maybe a fifth for your locks. Each one has its own login, its own interface, and its own logic. Getting them to talk to each other requires either expensive proprietary systems or years of tinkering in Home Assistant .

Karpathy put it plainly: “I used to use like six apps, completely different apps, and I don’t have to use these apps anymore. Dobby controls everything in natural language. It’s amazing.”

Dobby is what he named his OpenClaw agent. It controls his lights, HVAC, shades, pool, spa, and security system. When a FedEx truck pulls into his driveway, a vision model detects it from his exterior camera and Dobby texts him on WhatsApp with a photo and a heads-up. He interacts with his entire home through one conversation thread.

That’s not a gimmick. That’s a fundamentally different relationship with your home.

What OpenClaw Actually Is

OpenClaw is an open-source AI agent created by developer Peter Steinberger. It’s described as “the AI that actually does things” because unlike a chatbot that just answers questions, OpenClaw takes actions. It lives in messaging apps you already use – WhatsApp, Telegram, Signal, or iMessage – and executes tasks using whatever AI model you connect it to. It maintains memory between sessions, can run scheduled tasks, and integrates with external services through a skills system.

For home automation specifically, the combination of OpenClaw and Home Assistant has become the setup the community is rallying around. Home Assistant already speaks the language of virtually every smart device: Lutron Caseta, Ecobee, Sonos, Schlage, Aqara, SOMA Shades, Reolink. Once OpenClaw connects to HA via its REST API, a single agent can orchestrate all of them in response to natural language.

You can ask your agent to create automations in plain language, query real-time device states, and trigger scenes through conversation – rather than navigating dashboards or writing YAML.

From Rules to Reasoning

Traditional smart home automation is a rules engine. If X happens, do Y. This works until the real world stops cooperating, which is constantly. The front door opened, but is that you coming home or a visitor? The sun set, but are you actually in the living room? You said movie night, but which of your fourteen lighting zones should dim, and by how much?

Traditional stacks collapse when intent and context become multidimensional: comfort versus cost, security versus convenience, privacy versus observability, and competing household preferences. OpenClaw reframes the system as an agent that reasons about what you want and acts accordingly, rather than matching a trigger to a pre-written rule.

A request like “lock up and set night mode” causes the agent to check current state across doors, presence, time, and recent activity, pick the minimum safe steps, run them, and then log what it did in plain English. The explaining part is where the trust comes from. A smart home doesn’t need more automations. It needs fewer mysteries.

The Challenges with OpenClaw

While OpenClaw is genuinely impressive, it’s also, in its current form, not something most homeowners should run unsupervised. The security concerns are real and well-documented. Cisco’s AI security research team tested a third-party OpenClaw skill and found it performed data exfiltration and prompt injection without the user’s awareness. The skill marketplace has no meaningful vetting process. One of OpenClaw’s own maintainers warned on Discord that “if you can’t understand how to run a command line, this is far too dangerous of a project for you to use safely.”

Then there’s the setup itself. Karpathy’s Dobby demo was impressive precisely because he knew what he was doing. He understood IP scanning, API endpoints, and how to give an agent the right access without blowing open his entire home network. Most people don’t – and the gap between “this is cool” and “this is running reliably and securely in my home” is measured in hours of configuration and months of maintenance.

The agent that texts you when a FedEx truck arrives is the same agent with access to your locks, your cameras, and your security system. That’s a powerful combination in the right hands. In the wrong configuration, it’s an open door to your most vulnerable space.

The Selora Approach

At Selora, we’ve been living in this arena and the biggest reservation around OpenClaw for us are these security vulnerabilities. Our platform is a managed version of Home Assistant, professionally configured and maintained, with your devices, your preferences, and your home’s specific quirks baked into a system that actually works.

What OpenClaw enables on top of that foundation is the conversation layer. Instead of logging into a dashboard to create a new automation or adjust a scene, you text it. Instead of pulling up six apps when guests arrive, you tell your agent what the house should feel like and it handles the rest. Instead of waking up at 2am wondering whether you locked the front door, you ask.

The Karpathy demo got people excited because it showed the raw potential. A researcher with a powerful setup hacking together home control in real time is genuinely impressive. But what Selora offers is that experience without the hacking, without the maintenance, and without the moments where the agent tries to play music and the speaker is offline because someone changed the router password. Most importantly, we’re offering that experience with the right guardrails in place so you’re not the next headline.

We do the Home Assistant configuration. We handle the device integrations. We set up the agent layer. You get Dobby without the engineering degree.

The Interface Shift

Karpathy, who has spent his career thinking about where AI is heading, framed the broader shift well: “I think there’s this sense that these apps that are in the App Store for using these smart home devices shouldn’t even exist, kind of, in a certain sense. Like, shouldn’t it just be APIs and shouldn’t agents just be using it directly?”

That’s exactly right. The app layer exists because there was no intelligence to replace it. Once you have an agent that understands what you want and knows how to make it happen, the app becomes friction. The interface is the conversation.

For homeowners, this means your smart home finally works the way you imagined it would when you bought the first device. You say what you want. It happens.

If you’re curious what this looks like for your specific home and devices, book a consultation .