System configuration

Sensor data is transmitted from the microcontroller board via BLE UART. Since NemoClaw does not have a BLE UART interface, a sensor data server, which is a REST API endpoint, is set up, and NemoClaw retrieves data from that server.

Sensor data server

We created a sensor data server that publishes sensor data received via BLE UART using a REST API, and NemoClaw retrieves data from this server. Although the API is specific to this demo, it was easily implemented using FastAPI. The image below shows the created API displayed using Swagger UI.

Network Policy

The network policy for NemoClaw 's AI agent to access the sensor data server is as follows:

preset: name: sensor-api description: "Sensor API" network_policies: sensor_api: name: sensor_api endpoints: - host: xx.xx.xx.xx # 実際のIPアドレスを記載する port: 8100 protocol: rest enforcement: enforce rules: - allow: { method: GET, path: "/" } - allow: { method: GET, path: "/status" } - allow: { method: GET, path: "/sensor/latest" } - allow: { method: GET, path: "/sensor/history" } - allow: { method: GET, path: "/ble/scan" } - allow: { method: POST, path: "/ble/connect" } - allow: { method: POST, path: "/ble/disconnect" } binaries: - { path: /usr/bin/curl } - { path: /usr/local/bin/node }

Agent skills

The method for using the sensor data server is described in a format compliant with Agent Skills and installed using the `nemoclaw sandbox skill install` command. This agent skill allows the AI agent to understand how to acquire sensor data without prompting.

Example conversation with an AI agent

The following is an example of a request to an AI agent to read sensor data. "BME688" is the model name of the sensor device.

Example of a Cron job

The following is an example of using a Cron job to periodically monitor sensor data.

Prompt: "Refer to the bme688-sensor-api SKILL to monitor for any changes in the data output from the BME688 sensor."

Inquiry

In Episode 3, we explained how to periodically acquire temperature, humidity, and pressure sensor data and observe it using NemoClaw.
In Episode 4, we will explain how to use an acceleration unit (IMU) to detect changes in activity and notify an AI agent in real time.