macOS quietly built a knowledge graph of your life; iKnow reads it

Two undocumented macOS daemons map your contacts, routines and inferred relationships; an open tool now surfaces it. Plus Pika's open video+audio model.

Nowline NOV 8 9:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • macOS has been profiling you on-device

    A reverse-engineering deep-dive found two undocumented daemons — knowledgeconstructiond and intelligenceplatformd — building a triple-store graph in ~/Library/IntelligencePlatform: 181 entity classes, 278 predicates and 51 plaintext SQLite databases spanning contacts, locations, daily routines, Family Sharing ages and your Health Medical ID, with an XGBoost model inferring your family and social ties. The open-source iKnow dashboard (Apache-2.0, opens every database read-only) lets you see in a browser exactly what your Mac recorded — and what it merely guessed — which is worth knowing before you ship anything that trusts on-device data as private.

  • Open video with synced audio, in six steps

    Pika post-trained MiniMax H3 into a six-step model that generates 5–15s clips with synchronized 32kHz stereo audio, native at 480p/768p and up to 4K through its hosted API. Weights are on Hugging Face with ComfyUI and diffusers support, the API starts at $0.01 per output-second, and it can return separate Dialogue/Score/SFX/Foley/Ambience stems — enough to score a short film end-to-end without leaving your pipeline. It currently tops DesignArena's text-to-video board among open-weight models; note the MiniMax H3 Community License restricts commercial use, so check terms first.