feat: add Core AI provider package

README.md

@@ -45,11 +45,12 @@ receiving the current app's telemetry stream.
 
 ## Available Providers
 
-| Provider        | Built-in | Platforms    | Runtime                                                             | Description                                                |
-| --------------- | -------- | ------------ | ------------------------------------------------------------------- | ---------------------------------------------------------- |
-| [Apple](#apple) | ✅ Yes   | iOS          | [Apple](https://developer.apple.com/documentation/FoundationModels) | Apple Foundation Models, embeddings, transcription, speech |
-| [Llama](#llama) | ❌ No    | iOS, Android | [llama.rn](https://github.com/mybigday/llama.rn)                    | Run GGUF models via llama.rn                               |
-| [MLC](#mlc)     | ❌ No    | iOS, Android | [MLC LLM](https://github.com/mlc-ai/mlc-llm)                        | Run open-source LLMs via MLC runtime                       |
+| Provider            | Built-in | Platforms    | Runtime                                                             | Description                                                |
+| ------------------- | -------- | ------------ | ------------------------------------------------------------------- | ---------------------------------------------------------- |
+| [Apple](#apple)     | ✅ Yes   | iOS          | [Apple](https://developer.apple.com/documentation/FoundationModels) | Apple Foundation Models, embeddings, transcription, speech |
+| [Core AI](#core-ai) | ❌ No    | iOS          | [Core AI](https://developer.apple.com/core-ai/)                     | Run app-provided Core AI model bundles                     |
+| [Llama](#llama)     | ❌ No    | iOS, Android | [llama.rn](https://github.com/mybigday/llama.rn)                    | Run GGUF models via llama.rn                               |
+| [MLC](#mlc)         | ❌ No    | iOS, Android | [MLC LLM](https://github.com/mlc-ai/mlc-llm)                        | Run open-source LLMs via MLC runtime                       |
 
 ---
 
@@ -119,6 +120,47 @@ See the [Apple documentation](https://react-native-ai.dev/docs/apple/getting-sta
 
 ---
 
+### Core AI
+
+Run app-provided Apple Core AI model bundles exported from [`apple/coreai-models`](https://github.com/apple/coreai-models). This provider is for custom `.aimodel` assets, not Apple system models.
+
+- **Language Sessions** - Core AI language models with persistent native sessions
+- **AI SDK Adapters** - Language, embeddings, image generation, and transcription where AI SDK has matching interfaces
+- **Native-only APIs** - Segmentation, object detection, depth, super-resolution, model inspection, specialization, and raw `.aimodel` functions
+- **Catalog Metadata** - Starter model metadata from Apple’s registry, including iOS/macOS support
+
+#### Installation
+
+```bash
+npm install @react-native-ai/core-ai
+```
+
+Core AI also requires adding `https://github.com/apple/coreai-models` as a Swift Package dependency in the host iOS app and linking the products you use, such as `CoreAILM`, `CoreAIDiffusion`, `CoreAISegmentation`, and `CoreAIObjectDetection`.
+
+React Native installs the package through CocoaPods, but the Apple helpers are Swift Package products. Expo apps should use a config plugin to patch the generated Xcode project; bare React Native apps should add the package in Xcode or through a predictable project patch.
+
+#### Usage
+
+```typescript
+import { coreAI } from '@react-native-ai/core-ai'
+import { generateText } from 'ai'
+
+const model = coreAI.languageModel({
+  id: 'qwen3-0.6b',
+  source: { type: 'file', uri: qwen3ModelDirectory },
+  variant: 'iOS',
+})
+
+const { text } = await generateText({
+  model,
+  prompt: 'Explain Core AI',
+})
+```
+
+See the [Core AI documentation](https://react-native-ai.dev/docs/core-ai/getting-started) for setup, starter models, and direct native APIs.
+
+---
+
 ### Llama
 
 Run any GGUF model on-device using [llama.rn](https://github.com/mybigday/llama.rn). **Requires download** - models are downloaded from HuggingFace.

packages/core-ai/CoreAI.podspec

@@ -0,0 +1,21 @@
+require "json"
+
+package = JSON.parse(File.read(File.join(__dir__, "package.json")))
+
+Pod::Spec.new do |s|
+  s.name         = "ReactNativeAICoreAI"
+
+  s.version      = package["version"]
+  s.summary      = package["description"]
+  s.homepage     = package["homepage"]
+  s.license      = package["license"]
+  s.authors      = package["author"]
+
+  s.platforms    = { :ios => min_ios_version_supported }
+  s.source       = { :git => "https://github.com/callstackincubator/ai.git", :tag => "#{s.version}" }
+
+  s.source_files = "ios/**/*.{h,m,mm,swift}"
+  s.exclude_files = "ios/Tests/**/*"
+
+  install_modules_dependencies(s)
+end

packages/core-ai/README.md

@@ -0,0 +1,116 @@
+# @react-native-ai/core-ai
+
+Apple Core AI provider for React Native AI.
+
+This package is for app-provided Core AI model bundles exported from
+[`apple/coreai-models`](https://github.com/apple/coreai-models). It is separate
+from `@react-native-ai/apple`, which wraps Apple system capabilities such as
+Foundation Models, NLContextualEmbedding, SpeechAnalyzer, and AVSpeechSynthesizer.
+
+## API Layers
+
+Core AI exposes two layers:
+
+- AI SDK adapters where AI SDK has a matching model interface: language models,
+  image generation, text embeddings, and transcription.
+- Direct native APIs for Core AI-specific capabilities: model loading,
+  persistent language sessions, model inspection, specialization, segmentation,
+  object detection, depth, super-resolution, and raw `.aimodel` functions.
+
+The first native wrappers cover language sessions, diffusion image generation,
+segmentation, and object detection. Embeddings, transcription, depth,
+super-resolution, classification, reranking, speech generation, video
+generation, and arbitrary raw tensor calls stay explicit: embeddings and
+transcription keep AI SDK-shaped adapters because Apple has starter export
+recipes, while reranking, speech, and video have no provider helpers until a
+real Core AI model/runtime exists.
+
+## Native Setup
+
+Add `https://github.com/apple/coreai-models` as a Swift Package dependency in
+the host iOS app and link the products used by your wrappers:
+
+This is the intended integration path, but it still needs full verification in
+an Xcode 27 / iOS 27 host app with Apple Core AI products linked into the app
+target.
+
+- `CoreAILM` for language models and sessions
+- `CoreAIDiffusion` for image generation
+- `CoreAISegmentation` for segmentation
+- `CoreAIObjectDetection` for object detection
+
+React Native installs this package through CocoaPods, but Apple's helpers are
+Swift Package products. The app target still needs those SPM products linked.
+For Expo, add the config plugin. It links all known Apple Core AI Swift Package
+products by default:
+
+```json
+{
+  "expo": {
+    "plugins": ["@react-native-ai/core-ai"]
+  }
+}
+```
+
+Pass `products` only when you want to link a smaller set:
+
+```json
+{
+  "expo": {
+    "plugins": [
+      ["@react-native-ai/core-ai", { "products": ["CoreAILM"] }]
+    ]
+  }
+}
+```
+
+For bare React Native, add the Swift Package in Xcode or with a project patching
+script.
+
+## Usage
+
+```ts
+import { coreAI } from '@react-native-ai/core-ai'
+import { streamText } from 'ai'
+
+const model = coreAI.languageModel({
+  id: 'qwen3-0.6b',
+  source: { type: 'file', uri: qwen3ModelDirectory },
+  variant: 'iOS',
+})
+
+const result = streamText({
+  model,
+  prompt: 'Explain Core AI in one sentence.',
+})
+```
+
+For persistent native sessions:
+
+```ts
+await model.prepare()
+
+const session = await model.createSession({
+  instructions: 'Answer tersely.',
+})
+
+const response = await session.respond('What is Qwen3?')
+```
+
+For lower-level calls, import `CoreAI` and call the native TurboModule directly:
+
+```ts
+import { CoreAI, toNativeModelConfig } from '@react-native-ai/core-ai'
+
+await CoreAI.inspectModel(toNativeModelConfig(modelConfig))
+```
+
+## Model Management TBD
+
+Remote Core AI assets still need a first-class download path. Apple’s native
+API loads and specializes local `.aimodel` or `.aimodelc` URLs, so this package
+should add helpers that download model bundles into app storage, return a local
+file source, optionally specialize the model before first use, and persist
+bookmark data for cached specialized assets. Where possible, the downloader and
+storage pieces should be shared with the other React Native AI runtimes instead
+of duplicating provider-specific plumbing.

packages/core-ai/app.plugin.js

@@ -0,0 +1 @@
+module.exports = require('./lib/commonjs/expo-plugin')