macOS Sequoia adds on-device generative assistant APIs and privacy-first model hosting

AI · 6 min read

Apple announced that macOS Sequoia now exposes a system generative assistant API, allowing apps to request multimodal completions that respect system privacy settings and on-device policies. The assistant can run small to medium models locally while falling back to encrypted cloud operations only when users explicitly allow it.

To enable this, Apple added a hosted model tier inside macOS that supports quantized model formats and hardware-accelerated inference on Apple silicon. Developers can test model footprints in Xcode and take advantage of compiler-driven optimizations to balance latency and quality.

Apple also introduced privacy signals to govern prompt content and data retention, plus an attestation service for third-party models. The move signals a broader push to bring generative capabilities to native Mac apps while keeping personal data on-device by default.