A solar-powered greenhouse that thinks for itself.
234 sq ft · 4,979 ft elevation · Longmont, Colorado
Platform →
OpenClaw AI, ESP32, planning, data
Economics →
Solar, Powerwall, costs, efficiency
Equipment →
Every component with specs
Crops →
Profiles, VPD guide, DLI guide
Dashboards →
17 live dashboards, 10 perspectives
Code →
Source code & architecture
The Self-Reinforcing Loop
Measure → Store → Reason → Adjust → Execute
Every 5 seconds, the controller reads 6 climate probes and actuates equipment. Every 5 minutes, AI-planned setpoints push. Every 6 hours, the planner validates its decisions, extracts lessons, and writes the next plan.
The system learns by running — not by being told what to do.
The Stack
| Layer | Technology |
|---|---|
| Controller | ESP32 + ESPHome (42-state machine) |
| Agent | OpenClaw (Iris + Iris-dev) |
| Database | TimescaleDB (29 tables, 34 views) |
| Dashboards | Grafana OSS (17 dashboards) |
| Weather | Tempest + Open-Meteo |
| Energy | Shelly EM50 + Tesla Powerwall + Solar |
| Site | Quartz + Obsidian vault |
| Hosting | Docker + Traefik + Cloudflare |