# Apple Intelligence Apple's on-device and private-cloud AI stack, announced at WWDC 2024. Combines a small on-device foundation model running on the [[Apple Neural Engine]] with a larger model in Apple's "Private Cloud Compute" infrastructure for harder queries. Designed to be privacy-preserving by default. ## Architecture | Tier | Where | When Used | |---|---|---| | On-device model | iPhone, iPad, Mac (Apple Silicon) | Default for most queries | | Private Cloud Compute | Apple-operated, attestation-verified servers | Complex queries the local model can't handle | | Third-party (ChatGPT, etc.) | External providers, opt-in | When user explicitly invokes | The on-device model is reported to be ~3B parameters, distilled and quantized for the Neural Engine. ## What It Powers - Writing tools: rewrite, summarize, proofread (system-wide) - Image Playground: generate images on-device - Genmoji: custom emoji generation - Smart Reply, Notification summaries - Siri's natural language understanding (substantially upgraded) - Photos: cleanup, semantic search ## Privacy Model - Default execution is local; data never leaves device - For Private Cloud Compute, Apple publishes a verifiable attestation that no logs are kept and the code matches public images - Third-party model use (ChatGPT integration) requires explicit per-query consent This is a stronger guarantee than typical cloud AI services and a key marketing differentiator. ## Hardware Requirements Requires devices with sufficient NPU and RAM: - iPhone 15 Pro / Pro Max and newer - iPads with M-series chips - Macs with M-series chips - 8GB RAM minimum This effectively raised Apple's hardware floor, accelerating upgrade cycles. ## Relationship to Web Standards Apple has been quieter than Google on browser-based AI. If/when Safari implements [[WebNN API]] and [[Prompt API]], Apple Intelligence's on-device foundation model is the most plausible backing model — analogous to [[Gemini Nano]] in Chrome. ## References - https://www.apple.com/apple-intelligence/ ## Related - [[Apple Neural Engine]] - [[Apple Core ML]] - [[On-Device Machine Learning]] - [[Edge AI]] - [[Browser-Provided Language Models]] - [[Gemini Nano]] - [[Windows Copilot Runtime]] - [[Knowledge Distillation]] - [[AI Quantization]] - [[Small Language Models (SLMs)]] - [[AI Privacy]]