How to Run an AI IoT & MQTT Agent with OE Runtime

Subscribe to sensor feeds and publish commands to IoT devices via MQTT. OE Runtime lets you run a local AI agent that monitors device data, evaluates thresholds, and publishes automated responses — no IoT platform subscription required.

The Agent YAML

Create a file called agent.yaml inside an iot-messaging/ directory:

name: IoT Agent
description: Publish commands and read sensor data via MQTT
instructions: |
  You are an IoT agent. Subscribe to sensor data feeds and publish commands to devices via MQTT.
  Format all messages as JSON. Always read current state before publishing any commands.
  Complete all steps fully before writing your report.
steps:
  - name: Read Sensor Data
    content: |
      Subscribe to the sensor data topic and read the latest 5 messages.
      Extract: device ID, sensor type, value, unit, and timestamp for each message.
  - name: Evaluate and Publish Command
    content: |
      Analyze the sensor readings:
      - If any temperature reading exceeds 30°C, publish a cooling command to the control topic
      - If any humidity reading exceeds 80%, publish a ventilation command
      - Otherwise publish a status check ping to the control topic
      All commands must be JSON with fields: action, device_id, timestamp, triggered_by.
  - name: Report
    content: |
      Summarize IoT activity:
      - Sensor readings received (device, value, unit)
      - Any threshold breaches detected
      - Commands published (topic, payload)
      - Current system status (normal / alert)
connectors:
  - connection_name: MQTT Broker
    connection_type: mqtt

The Config File

Create oe-config.json in the same directory:

{
  "llm": {
    "provider": "openai",
    "model": "gpt-4o",
    "apiKey": "YOUR_OPENAI_API_KEY"
  },
  "server": {
    "enabled": false,
    "port": 3333,
    "apiKey": "your-secret-api-key"
  },
  "connectors": [
    {
      "connection_name": "MQTT Broker",
      "connection_type": "mqtt",
      "broker": "mqtt://localhost:1883",
      "clientId": "oe-runtime",
      "topic": "devices/#",
      "username": "YOUR_MQTT_USER",
      "password": "YOUR_MQTT_PASSWORD"
    }
  ]
}

Replace the broker URL with your MQTT broker address. This works with any MQTT broker including Mosquitto, HiveMQ, AWS IoT Core, and Azure IoT Hub. Remove username/password fields if your broker does not require authentication.

Download OE Runtime

Run the Agent

Windows

# Run the agent
oe-runtime-win.exe iot-messaging/agent.yaml --config iot-messaging/oe-config.json

macOS

# Make executable (first time only)
chmod +x oe-runtime-macos

# Run the agent
oe-runtime-macos iot-messaging/agent.yaml --config iot-messaging/oe-config.json

Linux

# Make executable (first time only)
chmod +x oe-runtime-linux

# Run the agent
oe-runtime-linux iot-messaging/agent.yaml --config iot-messaging/oe-config.json

Running as an API Server

Add --serve to start OE Runtime as an HTTP server on port 3333:

oe-runtime-win.exe --serve --config iot-messaging/oe-config.json

Then send a request:

curl -X POST http://localhost:3333/run \
  -H "Content-Type: application/json" \
  -d '{"yaml": "iot-messaging/agent.yaml", "params": {}}'

Or import the Postman collection (download above) to test all endpoints visually.

Use Cases

• Temperature and environment monitoring — Subscribe to temperature, humidity, and CO2 sensor feeds from smart building devices and automatically trigger ventilation, cooling, or heating commands when thresholds are breached. The agent evaluates multiple sensors simultaneously and publishes the appropriate response command without any manual intervention.
• Industrial equipment monitoring — Monitor vibration, pressure, and temperature sensors on manufacturing equipment and publish alerts or shutdown commands when readings indicate potential failure. Early detection of anomalies reduces unplanned downtime and extends equipment lifespan.
• Smart agriculture automation — Read soil moisture, light intensity, and nutrient level sensors across a field and publish irrigation or lighting commands to connected actuators based on crop requirements. The agent can be configured with crop-specific thresholds for different zones of the farm.
• Energy management — Subscribe to smart meter and sub-meter data to monitor real-time energy consumption across facilities. Publish load-shedding commands to non-critical devices when consumption exceeds a set threshold to reduce peak demand charges.
• Warehouse and cold chain monitoring — Monitor temperature and humidity in cold storage facilities and publish alerts when conditions drift outside safe ranges for stored goods. For pharmaceutical or food cold chains, the agent can also log readings to an audit trail for compliance purposes.
• Fleet and asset tracking — Subscribe to GPS and telemetry data published by vehicles or mobile assets and generate position reports or geofence alerts. The agent can identify when an asset leaves a defined zone and publish a notification command to the fleet management system.
• Security and access control — Read motion sensor, door contact, and access reader events via MQTT and publish lock, alarm, or notification commands based on the detected activity. The agent can apply time-based rules — for example, flagging motion detected outside business hours.
• Predictive maintenance scheduling — Analyse patterns in sensor data over time to identify equipment that is trending towards a failure threshold. The agent publishes a maintenance request event to the MQTT topic monitored by your work order management system before failure occurs.