RC RANDOM CHAOS

Every Mac's Hidden Daemons: Building a Personal Knowledge Graph

· via Hacker News

Original source

Every Mac comes with an (undocumented) family member classifier

Hacker News →

Every Mac running macOS 13 and later contains two daemons that continuously build a personal knowledge graph—a database mapping relationships, activities, and patterns across your digital life. The system aggregates data from Contacts, Calendar, Photos, Messages, Mail, location information, Screen Time, and behavioral signals into a triple-store database containing 181 entity classes and 278 predicates. The graph resides in plaintext SQLite databases under ~/Library/IntelligencePlatform, protected by macOS’s TCC full disk access controls.

Construction of the graph combines multiple inference approaches. Data directly imported from system apps forms the foundation, supplemented by machine learning models. Two XGBoost classifiers evaluate 144 behavioral features to infer family relationships, embedding models identify records belonging to the same person, image analyzers score relationship types (partner, parent, child, coworker, friend) from photo analysis, and the on-device LLM reads messages to assign names and relationships to faces. Each inferred relationship receives a confidence score between 0 and 1. The system recognizes fifty-seven categories of activity including sleeping, exercising, commuting, and dining.

The knowledge graph operates with minimal official documentation—only a sparse man page describes the knowledgeconstructiond daemon. This personal data collection system runs on both Intel and Apple silicon machines, though Apple has disclosed few details about the specific data elements gathered or inference mechanisms employed.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.