Apple unveils on-device generative AI framework for iOS and macOS

AI · 4 min read

At its latest developer release, Apple introduced a new on-device generative AI framework that standardizes model loading, inference, and safety filtering across iOS and macOS. The framework exposes a single API surface for text, vision, and audio generators and includes runtime optimizations for Apple silicon, including dynamic quantization and memory-friendliness for mobile devices.

Privacy is front and center: developers can mark models and pipelines as private by default, with system-managed differential privacy options and explicit user permission flows. Apple also includes a model manifest format so apps can declare provenance, versioning, and permissible uses; the system can block network transmission for models flagged as sensitive.

Designers and UX teams will get new tools in Xcode and Sketch-integration plugins to preview generative outputs inline and simulate user permission prompts. The move positions Apple to compete with cloud-first assistants by enabling richer, faster experiences with reduced latency and clearer user consent paths.