Clawrophyll: A Houseplant That Runs Its Own AI Agent on an Arduino UNO Q

Arduino® UNO™ Q 4GB

Arduino App Lab

<html><body><h3>Meet Phil</h3>He is a small Japanese cedar living in a 3D-printed crab-pot planter on my desk, and he runs his own AI agent on the Arduino UNO Q. He watches his soil, temperature, and light, shows how he feels through ten animated faces on a round display, and waves a servo pincer. When something matters, like his soil drying out, he messages me on Telegram. Say "Clawrophyll" out loud and he waves, checks his sensors, and tells you how he is doing, with his voice, out loud. He runs on a full agent with web access, a scheduler, and a camera, so he can set his own watering reminders, look up what light a cedar actually needs and compare it against his own light sensor, or read a handwritten note held up to his webcam. My favorite trick: show him a sign that says "if you can read this, show me your angry face" and his face snaps to angry. <figure><img src="https://projects.arduinocontent.cc/4588be67-bb95-4f6a-86cd-f8b0e0d809fd.gif"/><figcaption></figcaption></figure> <figure><iframe src="https://youtu.be/y0BJqXKGaN4"></iframe></figure> <h3>How it works</h3> The UNO Q is two computers on one board, and Phil uses both. The AI agent (the open-source nanobot framework, with Claude as the brain) runs on the Linux side, where there is RAM and a network. The STM32 microcontroller owns everything real-time: the sensors, the display, and the servo. The two halves are built and deployed together as one Arduino App Lab project, and the agent reaches the hardware through a small MCP server. The important part is that the firmware enforces hard limits, like how far the arm can swing, so the AI can never drive the hardware past a safe range no matter what it decides. <figure><img src="https://projects.arduinocontent.cc/5dc86b6b-cd3c-4949-aaa6-97ec4847a649.png"/><figcaption></figcaption></figure> The machine learning is tiered. A tiny keyword-spotting model, trained in Edge Impulse Studio, runs on the board around the clock and listens for the wake word "Clawrophyll" in about 17 milliseconds per inference, with no audio ever leaving the device. Only when it fires does the large cloud model wake up to reason, look through the webcam, and talk. Cheap and private at the edge, expensive and capable in the cloud, each where it belongs. <figure><img src="https://projects.arduinocontent.cc/7914d1bf-2698-49e1-b445-bf3269ed895c.jpg"/><figcaption></figcaption></figure> <figure><img src="https://projects.arduinocontent.cc/8bcc6409-a664-4041-8eda-99272d498bf9.gif"/><figcaption></figcaption></figure> The planter is four printed parts: the crab shell that holds the electronics, a water reservoir, the pot, and a stand for the sensors. Watering is built into the design: pour into the hole on the shell, the reservoir fills, and the pot draws water up from below, so no water ever goes near the boards. <figure><img src="https://projects.arduinocontent.cc/bf7240b1-d9c7-4a78-acd1-e42f64860ad4.png"/><figcaption></figcaption></figure> <figure><img src="https://projects.arduinocontent.cc/a3f89469-f9fe-44be-aa6f-1cb48b2e5701.png"/><figcaption></figcaption></figure> <figure><img src="https://projects.arduinocontent.cc/f0f3b7e5-4c5e-4b06-89b2-74ce57f79020.jpg"/><figcaption></figcaption></figure> <figure><img src="https://projects.arduinocontent.cc/681d5f71-0e86-43f8-8a41-e2a516d76758.jpg"/><figcaption></figcaption></figure> <figure><img src="https://projects.arduinocontent.cc/76164547-deb6-4f6b-b69b-aff4cc362b48.jpg"/><figcaption></figcaption></figure> <h3>Build your own</h3> This page is the short version. The full write-up, with the wiring diagram, bill of materials, every build step, the Edge Impulse training walkthrough, and a troubleshooting table, lives on Hackster ( <a href="https://www.hackster.io/s4muela/clawrophyll-a-houseplant-that-runs-its-own-ai-agent-4c21ef">https://www.hackster.io/s4muela/clawrophyll-a-houseplant-that-runs-its-own-ai-agent-4c21ef </a>), and all the source code, CAD files, and images are on GitHub ( <a href="https://github.com/SamuelAlexander/clawrophyll">https://github.com/SamuelAlexander/clawrophyll </a>). The trained wake-word model is a public Edge Impulse project you can clone ( <a href="https://studio.edgeimpulse.com/public/1031403/latest">https://studio.edgeimpulse.com/public/1031403/latest </a>). If you build a Phil of your own, I would like to see it. So would he, though he would never admit it. </body></html>

Clawrophyll: A Houseplant That Runs Its Own AI Agent on an Arduino UNO Q

A potted cedar with a face, a pincer, senses, and a voice: a full AI agent on the Arduino UNO Q, held inside firmware-enforced limits.