Privacy Infrastructure for Smart Glasses Build smart glasses apps without privacy concerns. Smart glasses apps face privacy hurdles. This real-time privacy filter sits between the camera and the app, automatically ensuring compliance. How it works: Replace your raw camera feed with our filtered stream. The filter processes live video, applies privacy protections, and outputs a compliant stream in real time. Use this processed stream for AI apps, social apps, or anything else. Features: Anonymizes faces of non-consenting individuals (configurable: blur or solid ellipse masking) of non-consenting individuals (configurable: blur or solid ellipse masking) Manages consent (e.g., detects verbal consent such as "I consent to be captured" and remembers it) (e.g., detects verbal consent such as "I consent to be captured" and remembers it) Real-time processing – 720p 30fps on laptop – 720p 30fps on laptop 100% offline – no cloud dependencies – no cloud dependencies Recording Developer-friendly: Easy camera replacement RTMP input / multiple output formats HTTP API for control Demo demo.mp4 Using a smartphone as the camera. Smart glasses demos here and here. Tech Stack Runs offline on a laptop. Built with FFmpeg (stream decode/encode), OpenCV (face detection/anonymization), Faster Whisper (voice transcription), and Phi-3.5 Mini (LLM for consent detection). Quick Start See the backend README, example app README, and CLAUDE.md for technical details. Use Cases Works with any camera-based app, for example: AI Assistants – Memory augmentation without privacy risks – Memory augmentation without privacy risks Social Apps – Live streaming with automatic protection – Live streaming with automatic protection Enterprise – Compliant workplace recording – Compliant workplace recording Content Creation – Automatic face anonymization for vlogs Roadmap Integration guides for various smart glasses Additional privacy filters (text, objects) Speech anonymization Location-based auto shutoff Legal compliance templates VLM integration Active speaker detection More input protocols (currently RTMP) etc. Contributing Feedback, questions, and contributions are welcome.