Beginner’s Guide: Building a Pedestrian Detection Skill with NanoBot

01 Introduction

The AI wave has sparked interest in lightweight agents. NanoBot is an ultra‑lightweight Python project (≈4,000 lines) that lets users create AI‑driven skills through natural language.

02 Install NanoBot

Create a Conda environment and install the package:

conda create -n claw python=3.12

conda activate claw

pip install nanobot-ai

Verify the installation: nanobot --version Successful output confirms the tool is ready.

03 NanoBot Configuration

Run the onboarding command to generate configuration files, workspace, and template documents: nanobot onboard Create config file ~/.nanobot/config.json Create workspace ~/.nanobot/workspace/ Generate template files AGENTS.md, SOUL.md, USER.md Edit ~/.nanobot/config.json to set the model name under agents and insert your DeepSeek API key and base URL as shown in the screenshots.

04 Verify Configuration

Test the setup with a simple command: nanobot agent -m "Hello" Correct output indicates the configuration works.

05 Project Practice – Pedestrian Detection Skill

Enter agent interaction mode: nanobot agent Issue the natural‑language request:

Implement a pedestrian detection SKILL, recommend using YOLO, function: user provides an image, detect pedestrians.

NanoBot generates the skill code in ~/.nanobot/workspace/skills. The main file pedestrian_detection.py contains fully commented, well‑styled Python code.

06 Code Testing

Place a test image and the YOLO model file ( yolov8n.pt) in a local folder, then run the generated skill:

nanobot agent -m "Test the pedestrian detection on C:\\Users\\zhaoq\\Desktop\\sample\\test.jpg using model C:\\Users\\zhaoq\\Desktop\\sample\\yolov8n.pt"

The tool processes the image, saves visualized detection results, and displays them as shown in the screenshots.

07 Function Extension – Export to JSON

Ask NanoBot to extend the skill to export prediction results as JSON:

Extend the above function to save predictions in JSON format.

Running the generated command creates a .json file in the output directory, confirming the extended capability.

08 Conclusion

The tutorial demonstrates how to install NanoBot, configure it, and use natural language to generate a complete pedestrian detection pipeline with YOLO, test the code, and add JSON export—all with minimal manual coding.