A security camera that sees and reasons, fully on-device

Detect people, vehicles, and animals, reason about risk with a vision-language model, and never send a frame off the machine.

git clone https://github.com/tetherto/qvac-examples
cd qvac-examples/qvac-smart-camera
npm install
npm start

QVAC Smart Camera is an illustrative example that runs object detection and scene reasoning entirely on-device, instead of streaming footage to a company's servers. YOLOv10-M draws color-coded boxes around people, vehicles, animals, and bags in real time; a Qwen3-VL 2B vision-language model then reasons about the scene to assess risk, combined with deterministic rules like brightness analysis for nighttime detection. Flagged events get an on-screen alert, a border flash, an alarm sound, and a logged entry with timestamp, thumbnail, and description.

Prerequisites

• Node.js 20 or later

• 8 GB RAM minimum, 16 GB recommended

• ~1.6 GB disk for cached models (YOLOv10-M + Qwen3-VL 2B)

• GPU optional - supported on Apple Silicon and Windows

• A webcam or camera feed to point the demo at

npx -y @qvac/cli doctor

git clone https://github.com/tetherto/qvac-examples
cd qvac-examples/qvac-smart-camera
npm install

npm start