AI Baby Monitor: A Local Video-LLM Powered Solution for Child Safety

AI Baby Monitor: A Local Video-LLM Powered Solution for Child Safety

Summary

The AI Baby Monitor is an innovative, privacy-first solution that leverages local Video-LLMs to enhance child supervision. It monitors a video stream against user-defined safety rules, issuing a gentle beep if a rule is broken. This tool acts as an additional pair of eyes, providing real-time alerts without compromising privacy.

Repository Info

Updated on February 22, 2026
View on GitHub

Tags

baby-monitorvideo-llmPythonAILLMprivacyhome-securityopen-source

Click on any tag to explore related repositories

Introduction

The AI Baby Monitor is a powerful, privacy-first project that utilizes local Video-LLMs to provide an intelligent monitoring system for children. Designed to be your "second pair of eyes," it watches a video stream from a webcam or RTSP camera and applies a simple list of user-defined safety rules. If a rule is broken, the system issues a single, gentle beep, prompting you to quickly check on your baby. Everything runs locally on your network, ensuring complete privacy.

Installation

Getting started with the AI Baby Monitor requires Docker, docker-compose, a GPU, and Python 3.12 with uv.

Prerequisites: Docker + docker-compose, One GPU, Python 3.12 with uv

# 1 — clone the repository
$ git clone https://github.com/zeenolife/ai-baby-monitor.git && cd ai-baby-monitor

# 2 — copy .env.template into .env
$ cp .env.template .env

# 3 — build & start all services (Redis, vLLM, video streamer, Streamlit viewer)
$ docker compose up --build -d

# 4 — start the watcher on the host. This is necessary for sound playback.
$ uv run scripts/run_watcher.py --config-file configs/living_room.yaml

# 5 — open the dashboard ? http://localhost:8501. You can also open the dashboard on your phone http://{host_network_ip}:8501

Please note, the first run will download the model (approximately 6 GB), build the Docker image, and may take a few minutes.

Examples

The project includes compelling demos that illustrate its functionality without putting children in danger.

  • "No smartphones" rule: A demo shows an alert being fired when people are using smartphones, violating a predefined rule.
  • Baby walking safely: Another demo illustrates the system correctly identifying a baby walking safely, with no alert issued.

These examples highlight the system's ability to discern between safe and potentially unsafe situations based on your custom rules.

Why Use It

The AI Baby Monitor stands out for several key reasons:

  • Private-first: All processing occurs locally, ensuring no data ever leaves your network. This commitment to privacy is paramount for sensitive monitoring applications.
  • Realtime-ish Performance: It operates efficiently on consumer GPUs, processing at approximately one request per second, making it practical for home use.
  • Advanced Video LLM: By default, it uses the Qwen2.5 VL model, served through vLLM, providing sophisticated video analysis capabilities.
  • Minimalist Alerts: A single, gentle beep serves as a deliberate, quiet notification, designed to prompt a quick glance rather than cause alarm.
  • Live Dashboard: A Streamlit viewer offers a live stream and real-time LLM reasoning logs, giving you full transparency into the system's operations.
  • Easy Configuration: Safety rules are defined in natural language within simple YAML files, making customization straightforward.
  • Multi-room Support: The system can be configured to monitor multiple rooms simultaneously, adapting to various home setups.

It is important to remember that this project is not a replacement for adult supervision. It is intended as an additional safeguard for those brief moments of distraction, providing a timely alert when needed.

Links