About
Giving my houseplants an agent.
Plantbot is a little experiment in agentic-fying inanimate objects. It's a Raspberry Pi hooked to my actual houseplants — sensors, pumps, a camera, the whole thing — then I give it Claude and tools so the plant agent can perform actions like telling you the moisture level of a plant, or watering it.
How the pieces fit together
A Raspberry Pi sits on the shelf next to the plants. It reads the moisture, humidity, and temperature sensors, drives a small pump for watering (coming soon), and constantly writes the data to my backend server. The camera streams the plants over HLS so you can actually watch them.
This website is the front-end. It pulls sensor data and the video feed from the Pi, and forwards your chat messages to Claude. Claude has a handful of tools wired up — read a sensor, water a plant, write back to the chat — and decides on its own when to call them. When it waters a plant, an actual pump runs; a few seconds later the moisture reading jumps and you can see the soil get darker on camera.
That's basically it. Building agent is very easy, and anyone can do it. There's no clever architecture, no orchestration framework — just a Pi, some cheap sensors, an LLM, and a bunch of glue code.
Why
I wanted to see what it feels like when an agent can actually do things in the physical world, even tiny things. Most “AI agents” I'd built before that point only moved bits around. Giving one a pump and a real plant that can die felt like a more honest test.
Open source
The code and the wiring are all on GitHub. If you want to build something similar, the Guide walks through the hardware and the agent setup.