Every Mac running macOS 13 or later ships with a pair of background daemons whose primary function is to build what Apple calls a Personal Knowledge Graph: a comprehensive digital profile containing information about who you are, who you know, how people relate to each other, communication patterns, locations visited, and activities performed.
According to analysis published on 8rank.ai, this system operates with minimal public documentation. The closest official reference is a Unix manual page for “knowledgeconstructiond,” offering scant technical details. The system applies to both Intel and Apple silicon Macs released since 2022.
The knowledge graph is constructed by cross-referencing data from multiple sources on the device: Contacts, Calendar, Photos (including facial recognition), Messages, Mail, location history, Screen Time data, and what researchers describe as “a long tail of behavioral signals.” The system stores its output in 51 SQLite databases located in ~/Library/IntelligencePlatform, protected by macOS’s Transparency, Consent, and Control (TCC) Full Disk Access permissions.
The core database, graph.db, functions as a triple store—a database format organizing information as subject-predicate-object relationships. Researchers identified 181 entity classes (such as person, place, event, organization) and 278 predicates that describe relationships between entities. The system infers connections including family relationships through data like Family Sharing membership and ages, communication frequency and timing between contacts, and physical proximity based on photos and contextual signals.
The Personal Knowledge Graph feeds into Apple Intelligence features, with the databases queryable by large language models running on the device. Apple’s own researchers published academic work on the underlying techniques in 2023, but public documentation regarding macOS integration, which versions are affected, what data is collected, and how relationship inference operates remains sparse.
Researchers emphasize that while the databases are stored in plaintext, they reside in a protected zone. They highlight potential privacy concerns if such a comprehensive personal knowledge base were accessed without authorization, particularly when combined with language models, describing it as a “privacy single point of failure.” The researchers have open-sourced a tool called iKnow allowing users to examine their own knowledge graph contents on their Mac.
Key facts
- Every Mac since macOS 13 (2022) runs two background daemons that build a Personal Knowledge Graph
- The system indexes data from Contacts, Calendar, Photos, Messages, Mail, location, Screen Time, and Family Sharing information
- The knowledge graph is stored in 51 SQLite databases using a 181-class entity model with 278 relationship predicates
- The system infers family relationships, communication patterns, and physical proximity between contacts
- Apple’s implementation lacks public documentation beyond a basic Unix manual page entry