How does the Content Filter work?

The Content Filter is an ULTIMATE feature that automatically detects adult or inappropriate content in the media files available on the monitored device. When content above the configured confidence threshold is detected, an instant push notification is sent to your phone via the AllTracker Control Center companion app.

How it works

AllTracker scans media files that are publicly accessible on the monitored device, such as photos and videos stored in the device's gallery, and passes each file through an on-device AI classification model. Files in private, hidden, or system-protected folders are never accessed. The entire classification process runs locally on the target device, so no image data leaves the device at this stage.

The AI model evaluates each media file and assigns one of five content categories:

• Neutral - everyday, safe-for-work content

• Drawings - illustrated or cartoon content classified as non-explicit

• Hentai - adult-themed animated or illustrated content

• Explicit - suggestive or provocative content that may not be appropriate for children

• 18+ - explicitly adult sexual content

A notification is triggered when the model's confidence score for any adult category exceeds the sensitivity threshold you set in the Control Center settings.

About the AI model

The classification model is based on MobileNetV2 - a lightweight convolutional neural network architecture specifically designed for mobile and embedded devices. It delivers an excellent balance between accuracy and computational efficiency, making real-time on-device inference practical even on mid-range Android hardware.

The model was trained on a curated dataset of hundreds of thousands of labeled images spanning all five categories. It is exported in TensorFlow Lite (.tflite) format so it runs fully offline, without requiring any internet connection or external API call.

Data privacy & storage

Privacy is a core design principle of the Content Filter. Here is exactly how your data is handled:

• On-device processing: All AI classification runs entirely on the monitored device. Media files are never automatically uploaded to AllTracker's servers.

• Live thumbnail preview: When your monitoring session is actively open in the browser, a thumbnail of a flagged event is streamed directly to your connected session via the socket server. This preview is never stored on AllTracker's servers.

• Upload on demand: Only when you click the thumbnail to view the full image is it uploaded to AllTracker's servers and displayed in your browser.

• Storage period: Any uploaded image is stored on our servers for a maximum of 30 days, after which it is automatically and permanently deleted.

• Manual deletion: You can delete any previously uploaded file at any time from your Profile page under the tab: files.