GreenhouseAgent

GreenhouseAgent is a greenhouse monitoring and control stack that combines Arduino sensor ingestion, a FastAPI control service, a lightweight dashboard, and model-driven recommendations.

It keeps the operating loop in one place. Read sensor data. Pull weather context. Suggest or issue control actions. Keep an operator-visible log of what happened.

System

• Arduino firmware reads temperature, humidity, and light
• server/serial_relay.py bridges the serial stream into the API
• server/main.py tracks live state, weather context, and actuator commands
• web/ provides the operator dashboard
• Online mode uses an OpenAI prompt path and local mode falls back to a Random Forest model

Project structure

• arduino/ firmware for the sensor and actuator loop
• server/main.py API server and greenhouse state manager
• server/serial_relay.py serial bridge between the Arduino and the API
• server/rf_recommendation.py offline recommendation path
• web/ dashboard for readings, actuator state, weather, and recommendations
• data/ sample readings and the shipped local model artifact

Setup

1. Install dependencies

python3 -m pip install -r requirements.txt

2. Flash arduino/greenhouse_uno/greenhouse_uno.ino to the Arduino

3. Create a .env file with the keys below

OPENAI_API_KEY=sk-...
SUPABASE_URL=https://...
SUPABASE_KEY=...
SERIAL_PORT=/dev/ttyUSB0

4. Start the API server

python3 server/main.py

5. Start the serial relay

python3 server/serial_relay.py

6. Open http://127.0.0.1:8000

Recommendation flow

• A live reading becomes the current greenhouse state
• Weather data is fetched for the configured coordinates
• Online mode builds a recommendation from sensor state and forecast context
• Local mode uses data/rf_model.joblib for offline fallback
• Commands are normalized before they are sent to the actuator queue

Hardware

• Arduino Uno or compatible board
• DHT temperature and humidity sensor
• Analog light sensor
• Two servos for vent and shade control

Limitations

• This is a single-greenhouse prototype
• Operator review still matters for watering and hardware safety
• The default weather coordinates point to Boston until changed in the dashboard