Apple Intelligence Foundation Server (macOS)

A Swift command-line HTTP server that exposes the Apple Intelligence system language model over a simple JSON REST API using Apple's FoundationModels framework. Run it locally on supported devices to perform on-device language model inference.

This package is intended for experimentation and local development with Apple Intelligence.

---

Requirements

• macOS 26.0+ (or iOS 26.0+, iPadOS 26.0+, visionOS 26.0+)
• Swift 6.2+
• Apple Intelligence enabled on your device
• Xcode 17+ (for building)

Only devices that support Apple Intelligence can use this server. You must enable Apple Intelligence in System Settings.

---

Project Structure

apple-intelligence-foundation-server/
├── Package.swift              # Swift Package Manager configuration
├── Sources/
│   └── App/
│       └── main.swift         # Server implementation
└── README.md

---

Installation

1. Clone or navigate to the project directory
2. Resolve dependencies:

   swift package resolve

---

Usage

Running the server

swift run

The server will start on:

http://localhost:8080

---

API Overview

All responses are JSON. Errors are also returned as JSON with a consistent shape:

{
  "error": "Human-readable error message"
}

Endpoint summary

Method	Path	Description
POST	/inference	Run text generation via Apple Intelligence
GET	/health	Basic liveness/health check

---

API Endpoints

POST /inference

Send a prompt and receive a generated response from Apple Intelligence.

Request:

curl -X POST http://localhost:8080/inference \
  -H "Content-Type: application/json" \
  -d '{"prompt": "What is Swift programming?"}'

Request body:

{
  "prompt": "Your prompt text here"
}

Successful response:

{
  "response": "Generated text from Apple Intelligence..."
}

If the model is not available, you will receive an error response describing the issue.

---

GET /health

Health check endpoint to verify the server is running.

Request:

curl http://localhost:8080/health

Response:

{
  "status": "ok"
}

---

Implementation Details

• Web framework: Vapor 4.89.0
• AI integration: FoundationModels framework (Apple's on-device language model)
• Architecture: Async/await with Actor-based inference service for concurrency safety
• Port: 8080 (default Vapor HTTP port)
• Context window: Up to 4,096 tokens per session (approximately 12,000-16,000 characters for English)

Model capabilities

The Apple Intelligence system language model excels at:

• Text generation
• Summarization
• Entity extraction
• Creative writing
• Classification

Not suitable for: Basic math, code generation, complex logical reasoning

---

Best Practices

Performance tips

• Keep prompts focused and specific
• Limit prompt length for faster responses
• Use phrases like "in three sentences" to get concise responses
• Create a new session for each independent request
• Stay within the 4,096 token context limit

---

Troubleshooting

Model not available

If you get "Model is unavailable" errors:

1. Device not eligible
   Your device may not support Apple Intelligence. Check Apple's compatibility list.

2. Apple Intelligence not enabled
   Go to System Settings → Apple Intelligence and enable it.

3. Model not ready
   The model may still be downloading. Wait a few minutes and try again.

Build errors

If you encounter module import errors:

1. Ensure you're running macOS 26.0+ or equivalent platform version
2. Verify you're using Xcode 17+
3. Re-resolve dependencies:

   swift package resolve

4. If necessary, clean and rebuild:

   swift package clean
   swift build

---

References

• FoundationModels Framework Documentation
• Generating content and performing tasks with Foundation Models
• Apple Intelligence

---

License

This project is licensed under the MIT License.
See the LICENSE file for details.