Agentic AI Plant Care System
A minimalist, local-first AI system that autonomously senses, reasons, and acts to care for plants — powered by Jetson Nano, ESP32, and a local LLM agent.
A minimalist, local-first AI system that autonomously senses, reasons, and acts to care for plants — powered by Jetson Nano, ESP32, and a local LLM agent.
A true agentic system that senses, reasons, and acts autonomously using a local LLM — no hardcoded thresholds.
Runs entirely offline with local inference on Jetson Nano. No cloud, no internet — ideal for privacy and reliability.
When connected, the same system supports IoT dashboards, remote monitoring, logging, and cloud integration via MQTT.
All the hardware components required to build this agentic AI plant care system.
AI Processing Unit
Local LLM inference, MQTT broker, and agent logic
Edge Sensor Node
WiFi/MQTT enabled microcontroller for sensor readings
Soil Monitoring
Measures soil moisture level (analog output)
Environment Monitoring
Temperature and humidity sensor
Switching Control
Controls high-power water pump via low-power signal
Water Delivery
12V DC submersible pump for plant watering
Power Distribution
5V for Jetson/ESP32, 12V for water pump
Assembly
Jumper wires, breadboard, and connectors
Connectivity (Optional)
For MQTT communication and remote monitoring
The system is designed as a three-layer agentic architecture: perception at the edge, communication via MQTT, and intelligence powered by a local LLM running on Jetson Nano.
ESP32 continuously reads soil moisture and environmental data, then publishes structured telemetry via MQTT.
The Jetson-hosted LLM agent evaluates sensor data using domain knowledge and deterministic constraints.
Based on the agent’s decision, a precise watering command is sent to the relay-controlled pump.
A short video demonstrating real-time sensor input, agent reasoning, and autonomous watering actions.
Full source code including ESP32 firmware, Python agent logic, MQTT schemas, and LLM configuration.