Your Old Android Phone Is a Better Security Camera Than You Think
You've probably got a drawer full of retired Android phones—screens cracked, batteries tired, but the cameras still work fine. Most of them just sit there collecting dust while you consider buying dedicated IP cameras. Sentinel takes a different approach: it turns those idle phones into distributed camera nodes on your local network, with a PC dashboard handling the streaming and optional AI analysis.
What It Does
Sentinel is a real-time vision system built for local area networks. It splits the work across two pieces: CamFlow, an Android client that captures your phone's camera feed and uploads it as single-frame JPEG images, and a PC Dashboard (Flask plus a web UI) that receives those frames and gives you live preview, video recording, screenshot saving, log viewing, and optional AI-triggered monitoring.
The architecture is deliberately layered—"mobile capture + PC processing + browser control." Everything runs on your LAN without touching cloud services, though the README notes that if you want to run multimodal models, an online inference service is the recommended route. You can treat it as a lightweight local monitoring setup or as a prototype platform for visual data acquisition and intelligent analysis. The Android side requires version 8.0 or higher, and the PC side runs on Python 3.9+.
Why It's Cool
-
It solves a real hardware problem with hardware you already own. Dedicated cameras cost money and often phone home to a cloud you don't control. Sentinel repurposes what's already in your junk drawer and keeps the data on your network. That's a solid trade.
-
The separation of concerns is sensible. Phones are good at capturing images and bad at sustained compute. PCs are the opposite. By offloading frames to a PC for processing, you get real-time streaming and recording without asking a mid-range phone to do heavy lifting. The browser-based dashboard means you can watch from any device on the LAN.
-
AI monitoring is optional, not mandatory. The README describes AI-triggered monitoring with "controllable cognitive output" and structured event analysis, but the core system works without it. You can start with plain video preview and recording, then layer in analysis later. There's also a documented path for third-party model integration if you want to bring your own.
-
It's been noticed. The project was featured in ruanyf's weekly and xuanli199's Tech Review—not a guarantee of quality, but a sign that people are paying attention.
-
Honest about its constraints. The README includes a section on network quality and troubleshooting, which tells you the authors know LAN video streaming has real-world failure modes. That kind of documentation usually means someone actually ran this thing.
How to Try It
The README strongly recommends reading the documentation in order before your first run: project overview, then deployment, then running the project, then the dashboard guide. That's worth following—this isn't a single binary you double-click.
- Grab the source from the repository.
- Set up the PC side. You'll need Python 3.9 or newer. The README's deployment section covers the Flask dashboard setup.
- Install CamFlow on your Android device (8.0+). This is the capture client that streams frames to your PC.
- Start the PC dashboard, then launch CamFlow and point it at your server.
- Open the dashboard in your browser for live preview, recording, and screenshots. If you want AI monitoring, the README has a dedicated guide for the dashboard's AI module and third-party model integration.
The current release is v1.1.6, and there's a roadmap section in the README if you want to see where things are headed. Data locations are also documented, so you'll know where recordings and logs end up.
Final Thoughts
Sentinel is best suited to developers who are comfortable with a multi-part setup—an Android app plus a Python server—and who want a hackable foundation rather than a finished consumer product. If you just want plug-and-play surveillance, this will feel like work. But if you've got spare phones, a home LAN, and an interest in experimenting with local vision pipelines or multimodal AI, it's a genuinely practical starting point. The pattern it demonstrates—cheap sensors feeding a central processor—is one you'll see more of, and this is a reasonable way to learn it.
Follow @githubprojects for more developer tools and open source projects.