EyesOff

Product Introduction

1. EyesOff is a privacy-centric macOS application that actively monitors screen visibility using on-device machine learning. The app employs the YuNet convolutional neural network running locally to detect faces within your webcam's field of view in real time. It triggers immediate alerts when potential visual intrusions occur and stores encrypted snapshots of detection events without requiring internet connectivity. All processing occurs entirely on your Mac, ensuring no facial data or images leave the device.

2. The core value of EyesOff lies in its ability to protect visual privacy through edge-computed AI without compromising system performance or data security. By eliminating cloud dependencies and implementing hardware-optimized neural networks, it provides a zero-trust solution for sensitive work environments. The app operates as a lightweight background service, typically using less than 5% CPU load while maintaining continuous monitoring. This combination of robust privacy enforcement and efficient resource management makes it ideal for professionals handling confidential information.

Main Features

1. EyesOff utilizes YuNet, a lightweight face detection model optimized for Apple Silicon, achieving 15-30 FPS processing on M1/M2 chips while consuming under 100MB RAM. The model analyzes webcam input through adaptive frame sampling, dynamically adjusting detection frequency based on motion levels to balance accuracy and performance. Users receive instant desktop notifications and optional audio alerts when faces are detected within a configurable distance threshold.

2. All detection events trigger encrypted JPEG snapshots stored in the ~/.eyesoff/snapshots directory with timestamped filenames and EXIF metadata. Snapshots are protected using AES-256 encryption tied to the macOS Keychain, accessible only through user-authenticated sessions. The app automatically purges older snapshots based on customizable retention policies (default: 7 days) to manage storage efficiently.

3. The entire detection pipeline operates offline, with YuNet’s 1.2MB quantized model embedded directly in the application bundle. Network access is fully disabled at runtime through macOS sandboxing rules, and the app verifies its air-gapped operation via kernel-level socket monitoring. Compatibility extends to all macOS-native cameras and USB webcams supporting H.264 encoding at 720p resolution or higher.

Problems Solved

1. EyesOff addresses the vulnerability of screen visibility in open environments where physical privacy filters are impractical. Traditional solutions like manual screen dimming or position adjustments fail against determined observers, while cloud-based monitoring tools introduce data exposure risks. The app provides automated, real-time intrusion detection without requiring behavioral changes from users.

2. The primary user base includes remote workers handling sensitive data in public spaces, healthcare professionals managing patient records, and developers working with proprietary codebases. Government personnel requiring compliance with visual data protection standards (e.g., NIST SP 800-53) also benefit from its local processing architecture.

3. Typical scenarios include detecting shoulder-surfing attempts during financial spreadsheet analysis in coffee shops, preventing unauthorized viewing of confidential legal documents in co-working spaces, and securing coding sessions involving unreleased software features. The app proves equally valuable in home offices where multiple users share devices.

Unique Advantages

1. Unlike cloud-dependent alternatives like OBS Studio plugins or SaaS surveillance tools, EyesOff implements full edge computing with T2 Security Chip integration for model execution on Apple Silicon. This eliminates latency (under 50ms inference time) and attack surfaces associated with data transmission. Competitors’ solutions often require recurring subscriptions, while EyesOff uses a one-time purchase model.

2. The app introduces frame-differential processing that skips identical consecutive frames, reducing CPU usage by 30-40% compared to constant analysis. A proprietary background prioritization algorithm ensures detection processes yield resources to active applications during high system load. These innovations enable 24/7 operation without noticeable performance impact.

3. Competitive strengths include macOS-specific optimizations like Metal acceleration for neural networks and Touch ID-gated access to settings. Independent benchmarks show 98.6% detection accuracy in office lighting conditions, outperforming open-source alternatives like OpenCV’s Haar cascades. The app also supports headless operation for enterprise MDM deployments, a feature absent in consumer-grade tools.

Frequently Asked Questions (FAQ)

1. How does EyesOff handle false positives from photos or screensavers?
   The YuNet model is trained to ignore 2D facial representations through adversarial training techniques, reducing false positives from screens or printed photos by 92%. Users can further adjust detection confidence thresholds (default: 75%) via the configuration file. The app also implements motion validation checks requiring sustained face detection for 500ms before triggering alerts.

2. Can EyesOff operate when my Mac is in clamshell mode or connected to external monitors?
   The app automatically switches to external webcams when the laptop lid is closed, provided peripherals are connected before activation. For multi-monitor setups, it supports up to three concurrent camera inputs with individual detection zones mapped to specific displays. Users must enable "Allow EyesOff to monitor external devices" in Security & Privacy settings.

3. What happens to detection data if I uninstall the app?
   Uninstallation via the provided script permanently deletes the ~/.eyesoff directory, including all configuration files and encrypted snapshots. The neural network model and any temporary processing frames stored in macOS shared memory are purged during reboot. No residual data remains after complete removal.