Apple secretly built a personal knowledge graph into every Mac since 2022

Your Mac has been quietly mapping your relationships, habits, and locations into a plaintext database for two years.

Research · Source: Hacker News

What happened

Every Mac running macOS 13 or later ships with two hidden daemons named knowledgeconstructiond and intelligenceplatformd. Their only purpose in life is to build a massive graph of who you are and what you do. They map out who you know, how those people relate to each other, and how you communicate with them. This system has been running quietly in the background since 2022.

The daemons pull data from absolutely everywhere the operating system allows. They read your Contacts, Calendar, Photos, Messages, and Mail. They track your location, your Screen Time, and your Family Sharing details. They even log ambient light, Bluetooth connections, and when your screen is locked. All this information gets dumped into a directory of fifty-one SQLite databases stored in plaintext.

Researchers Dario Pasquini and Michał Bazyli just published a deep dive into this undocumented system. They found the main database is a triple store utilizing 181 entity classes and 278 relationship predicates. To help users understand what is being tracked, the researchers released an open-source local dashboard called iKnow. This tool lets anyone peek into the exact graph Apple has built about their household.

Key facts

Why it matters

Apple is building the ultimate context engine right under our noses. While independent developers struggle to build user context from scratch, Apple already possesses a structured, queryable database of every user's entire life. This graph is the hidden foundation for Apple Intelligence. It proves that the future of AI is local, highly personalized, and deeply integrated into the operating system itself.

This architecture creates a massive privacy single point of failure. The databases sit in plaintext, protected only by standard Full Disk Access controls. If malware or an unrestricted local LLM gets access to this specific directory, it gets a complete map of your family, your daily habits, and your physical movements. Security researchers are already running thought experiments on how to weaponize this data, proving that local storage does not automatically guarantee safety.

For builders

Local context is the ultimate moat

Apple does not need to ask users for context because they already have a structured graph of their entire life. Builders relying on manual user onboarding will lose to platforms with native OS-level access. You must find creative ways to build context without friction.

Plaintext storage is a massive risk

The graph databases sit in plaintext behind basic disk permissions. If you build local AI tools, you must encrypt your context stores. Users pay the ultimate price when local data gets exposed to malicious agents or rogue language models.

Study the open source forensics tools

The researchers released the iKnow dashboard to read this specific graph. Builders can study these schemas to understand exactly how Apple structures personal data and relationships. This is a free masterclass in building local knowledge graphs for your own AI products.

My take

I am not surprised Apple is hoarding this data to power their AI, but leaving fifty-one databases of personal relationships in plaintext is incredibly reckless. We spend all our time worrying about cloud privacy while our local machines quietly build a surveillance graph of our families. If you want to build AI products that people actually trust, you have to secure the data by default, not just hide the documentation from your users.

Original reporting: Hacker News. This is my rewrite and opinion.

More AI news for builders