Roof windows that open with the heat and slam shut before a storm. Watering by relay. A daily AI read on what the plants went through. Built at home, wired by hand, documented honestly — bugs included.
The real dashboard shows this scene in realtime over MQTT: two roof flaps that actually move, a climate map, and — since this week — a water-zone glow that honestly covers only the north bed, because that's the only bed valve 1 reaches.
Once a day the server sends 24 hours of temperature, humidity and cloud-cover history to Claude, which returns a short plain-language analysis: how the climate behaved, and early warnings for heat stress or mildew risk. It shows up on the dashboard like a note from someone who was watching all day.
“The indoor temperature remained stable overnight and in the early morning, with a slight drop to around 21°C near 03:00, but then rose quickly from 07:00, peaking at 26.7°C around 08:00. This peak is just within the acceptable upper limit for heat stress for the frost-sensitive crops. Despite full cloud cover during the day limiting solar radiation, relative humidity averaged high (71%), dropping to around 60% in the late afternoon during the highest temperatures but rising quickly in the evening and night to above 75% as temperatures fell. This creates a humid and cool climate, especially at night, which could increase the risk of fungal diseases like powdery mildew, while the daytime combination of warmth and humidity stayed within acceptable limits. Watch: check crops extra carefully in the coming days for early signs of powdery mildew or botrytis, especially in densely vegetated areas. Advice: consider manually increasing air circulation in the evening and night, if possible, to lower humidity somewhat, without causing too large a temperature drop.”
Every node is a €5 ESP32 talking MQTT to a small server. Fail-safe by design: lose the network and the windows close, valves shut themselves off, nothing floods.
Two nodes drive 12V linear actuators through BTS7960 H-bridges — one per roof flap, north and south. A weather guard on Open-Meteo data closes both windows on thunderstorms, gusts over 50 km/h or rain, then restores the last wanted position when it clears.
A sensor node publishes inside temperature and humidity over MQTT. The climate engine vents on hysteresis — open above 25°C, close under 21°C — with the weather guard always taking priority. At midnight, Claude analyses the full day.
A fail-safe-closed 1″ solenoid valve on a Grove relay, on its own node with its own firmware — deliberately decoupled from the climate logic. Manual for now, from a dashboard button or a shell script; soil-moisture automation is the next chapter.
The roof limit switches are probably dead or miswired, so the move-timeout is parked at 90 seconds as a workaround. The DS18B20 outdoor sensors never worked despite hours of diagnosis; a DHT11 does the job for now. That's what building real hardware looks like — and it all still runs.
The greenhouse control system graduated from simple monitoring to an actual working machine. We wired up the physical world: a dedicated ESP32 now controls the irrigation solenoid valve, and the window motor firmware got smarter with failsafes and diagnostic logging. The server can now trigger a full AI climate analysis on demand, which gives weekly optimization suggestions—think 'water less in the morning' rather than raw data. The dashboard became a proper control room with a permanent weather map, clearer logs, and even a basic 3D digital twin of the whole setup.
Firmware flashed and tested over USB; the relay clicks on command. Confirmed this week: valve 1 only waters the north bed — so the dashboard now honestly shows the water glow on the north plants only. Next up: wiring the 12/24V supply to prove the valve really opens.
climateControl.ts vents automatically on inside temperature with clean hysteresis. And once a day, the last 24 hours of climate history goes to Claude for a short written analysis with heat-stress and mildew warnings.
If enough people want it, two things could exist. Tell me which one you'd use — that's genuinely what decides the roadmap.
Pre-flashed ESP32 nodes, actuator drivers, relay board and wiring guide — retrofit any hobby greenhouse with self-venting windows and watering in a weekend. No soldering marathon, no firmware rabbit hole.
The dashboard and climate engine as a product: live scene of your greenhouse, weather-aware failsafes, and the nightly AI analysis — for growers who already have the hardware but not the picture.
Learn to build it yourself: a hands-on course from the real build log — wiring ESP32s and actuators, MQTT, fail-safe design, and hooking up the AI analysis. As a video series, workshop or a day in the actual greenhouse.
Leave your email and what you're interested in. No spam — just build updates and a first shout when the kit or software becomes real.