Apple Intelligence APIs and Tools for iOS 18

With the rollout of iOS 18, Apple Intelligence brings generative models directly into the operating system, exposing a dedicated set of APIs, frameworks, and enhancements that let developers integrate AI capabilities into their apps. Rather than requiring developers to host heavy server-side large language models, iOS 18 allows third-party software to leverage on-device intelligence for understanding context, generating media, refining text, and executing complex voice interactions. This article details the specific developer tools and APIs provided to access Apple Intelligence features.

App Intents and Siri Integration

The App Intents framework serves as the primary gateway for Apple Intelligence to interact with third-party apps. In iOS 18, Apple expanded App Intents to give Siri semantic indexing and on-screen awareness.

Developers can adopt predefined system schemas across categories such as photo editing, mail, document management, and messaging. By conforming to these schemas, apps expose specific actions and content that Siri can orchestrate. When a user asks Siri to perform an action—such as applying a specific photo filter or finding an attachment in a work app—the system relies on these structured App Intents to locate content and execute tasks across the operating system.

Writing Tools Integration

Writing Tools offers proofreading, text rewriting, and summarization across the system. For most applications, standard text views (UITextView, NSTextView, and WKWebView) inherit Writing Tools automatically with no code changes required.

For custom text implementations, Apple provides dedicated APIs within UIKit and AppKit to control or customize the behavior:

  • isWritingToolsActive: Detects when the user is actively using the tool, allowing the app to pause background operations or UI updates that might conflict with text replacement.
  • writingToolsBehavior: Allows developers to adjust the experience by enabling a complete inline editing panel, a limited non-inline panel, or opting out completely for fields requiring strict input validation (such as code editors or terminal views).
  • writingToolsResultOptions: Specifies whether the text view accepts plain text, rich text, or specific format types when transformations are applied.

Image Playground API

The Image Playground API allows apps to embed Apple’s generative image creation interface directly into their UI. Instead of creating a custom image generation pipeline, developers can present a native system sheet using UIKit (ImagePlaygroundViewController) or SwiftUI modifiers.

Key capabilities include:

  • Context Injection: Apps can seed the generator with concepts, text descriptions, or existing user images to create stylistically aligned visuals (such as illustrations or sketches).
  • Result Handling: Once the user finishes generating an asset, the API returns a standard image object directly to the app for immediate display or storage.

Genmoji and Rich Text APIs

iOS 18 supports Genmoji—user-generated contextual emoji—within standard text views. To ensure apps can display Genmoji alongside standard Unicode characters, Apple updated text formatting controls in NSAttributedString.

Apps using standard system controls display Genmoji automatically. For custom text engines, developers must adopt the latest rich text APIs that treat Genmoji as inline graphic attributes rather than standard glyphs, ensuring correct line heights, layout metrics, and copy-paste behavior.

Core ML and Foundation Models

Apple Intelligence operates using on-device Foundation Models. To complement this, Core ML has been updated to run alongside system models without exhausting device memory:

  • Stateful Model Support: Improves the performance and latency of custom large language models and transformers by maintaining cache states on the Apple Neural Engine (ANE).
  • Advanced Quantization: Tools in coremltools now support 2-bit to 8-bit weight compression, enabling developers to fit custom supplementary models alongside Apple's built-in models efficiently.