AI Agents & Wearables
Karpathy Built an AI Claw for His Home — We're Building One for Your Body
Andrej Karpathy's “Dobby the Elf Claw” proves AI agents can run your entire home. The next frontier? An AI Claw that understands your body.
Last week, Andrej Karpathy — former Tesla AI director and founder of Eureka Labs — shared something that made the tech world stop scrolling. He built an AI agent called “Dobby the Elf Claw” that controls his entire home: lights, HVAC, window shades, pool, spa, security cameras — all through natural language. Six separate apps, collapsed into one AI conversation.
He says “Dobby, it's sleepy time” and every light in the house goes dark. A package arrives, and the AI sends a WhatsApp notification with the security camera image. No app switching. No dashboards. Just one intelligent agent that understands his world.
This is the “Claw” paradigm — an autonomous AI agent that doesn't just answer questions but actually operates your devices, apps, and environment. And it's about to change everything.
1. What is an “AI Claw”?
The term “Claw” in the AI world refers to an autonomous agent that can reach into applications, devices, and services to take action on your behalf. Unlike a chatbot that only talks, a Claw does things.
Traditional Chatbot
“Your living room light is on. Would you like me to turn it off?” (You still open the app and tap the button.)
AI Claw
“Dobby, it's sleepy time.” (Every light turns off. Thermostat adjusts. Shades close. Done.)
Karpathy's system even runs on a local network — no cloud dependency — with an Nvidia DGX Station powered by a GB300 superchip for agentic AI workloads. This isn't a toy demo. It's production-grade autonomous control.
2. Home automation was the easy part
Controlling lights and thermostats is impressive — but it's fundamentally simple. These devices have binary or numeric states: on/off, 72°F, open/closed. The real challenge? Environments with ambiguous, continuous, deeply personal data.
Like your body.
Your wearable devices generate a constant stream of biometric data — heart rate variability, sleep stages, blood oxygen, strain scores, respiratory rate, body temperature. This data is more complex, more personal, and more valuable than any smart home metric. And right now, no AI Claw is touching it.
Karpathy consolidated six home apps into one AI. Most people have three or more health apps — Oura, Whoop, Apple Watch, Fitbit, Garmin — each locked in its own silo. The same fragmentation problem, but the stakes are your health, not your thermostat.
3. The wearable Claw: from smart home to smart body
If Karpathy's Dobby is a Claw for your house, what would a Claw for your body look like?
Karpathy's Home Claw
- • Controls lights, HVAC, shades
- • Monitors security cameras
- • Sends delivery notifications
- • Voice-activated commands
- • Local network, no cloud
MimiClaw's Body Claw
- • Reads Oura, Whoop, Apple Watch
- • Correlates sleep, strain, recovery
- • Proactive health recommendations
- • Natural language interface
- • Local-first, privacy-preserving
The pattern is identical: multiple siloed devices → one AI agent → unified intelligence → autonomous action. Karpathy proved it works for the home. The next step is proving it works for the body.
4. What a body-aware AI Claw actually does
Karpathy says “sleepy time” and his house responds. Imagine saying the same to an AI that knows your biometrics:
You: “Should I train hard today?”
MimiClaw: “Your Oura shows 5.8 hours of sleep with only 1h 10m deep sleep — well below your average. HRV is at 31ms (you usually sit around 45ms). Whoop recovery score is 34%. I'd recommend a mobility session instead. Your body typically bounces back after two good nights — aim for 10pm tonight.”
You: “Why have I been so tired this week?”
MimiClaw: “Looking at the last 7 days: your average sleep dropped from 7.4h to 6.1h. Body temperature has been 0.3°C above baseline since Tuesday. Respiratory rate ticked up from 14.2 to 15.8 breaths/min. This pattern often precedes getting sick. Consider resting today — and I've already dimmed your evening lights to start winding down at 9pm.”
This isn't science fiction. Every data point in those examples already exists in your wearables today. The missing piece is the Claw — an AI agent that reads all of it, connects the dots, and takes action.
5. Why “Claw” matters more than “chatbot”
The distinction Karpathy's work highlights is crucial. A chatbot answers questions. A Claw operates your world. The difference:
- Agency — a Claw doesn't wait to be asked. It notices your HRV crashing and adjusts your environment before you feel it.
- Cross-system reach — a Claw connects wearables, smart home, calendar, and health data into one unified graph.
- Continuous context — a Claw remembers that you slept badly Monday, trained hard Tuesday, and has been tracking your recovery since.
- Action, not advice — a Claw dims your lights, blocks your calendar, adjusts your thermostat. It doesn't just suggest — it executes.
This is exactly what Karpathy built for his home. And it's exactly what's missing for personal health. Your Oura app will never dim your lights. Your Whoop app will never reschedule your workout. But a Claw can — because it has access to everything and the authority to act.
6. Privacy-first: the non-negotiable
Karpathy made a deliberate choice: his AI Claw runs on a local network, not the cloud. For home automation, that's smart. For health data, it's mandatory.
An AI that knows your heart rate, sleep patterns, stress levels, and biometric trends knows more about you than your doctor. This data cannot live on someone else's server.
Local-first
All processing happens on your device. Your biometric data never leaves your hardware.
Open source
Every line of code is auditable. Trust is built on transparency, not promises.
Zero sharing
Your health data is never sold, shared, or used for advertising. Period.
Encrypted sync
When data moves between your devices, it moves encrypted end-to-end.
7. MimiClaw: the wearable Claw
MimiClaw started as an open-source AI agent running on a $5 microcontroller — proving that anyone could control hardware through natural language. We called it a “Claw” before the term went mainstream.
Now we're taking that same approach — one AI agent, one conversation, total device control — and applying it to the most personal data layer that exists: your body. One Claw that connects your Oura, your Whoop, your Apple Watch, your smart home, your calendar. One AI that truly knows you.
Karpathy proved the Claw paradigm works. He showed that one AI agent can replace six apps and make your environment intelligent. We're proving the same thing — but the environment is you.
The bottom line
Andrej Karpathy showed the world what an AI Claw looks like for the home.
The next frontier isn't another smart home agent. It's an AI that understands your body — across every wearable, every metric, every day.
The future of personal AI isn't a chatbot. It's a Claw that knows you better than you know yourself.
MimiClaw — the AI Claw for your body