@pxltail/apple-intelligence v1.0.4-beta

@pxltail/apple-intelligence

Version 1.0.4-beta

This is a basic library that lets you access Apple's new FoundationModels API(https://developer.apple.com/documentation/foundationmodels) via REST.

As of writing, it's the only library that does so.

Install

Installing this works like any other Node.js library.

npm install @pxltail/apple-intelligence

This will install it in the project you are currently in.

If you want to install this system wide to use the CLI tool anywhere:

npm install @pxltail/apple-intelligence -g

Requirements To Use

• macOS 26 Tahoe or later
• A Mac with an Apple Silicon processor (M1 or later)
• Apple Intelligence enabled, downloaded and installed
• Game mode disabled

Setting Up

Starting the server is very simple. There are two options, depending on how you're using the library.

• If you're using this in a Node.js script:

import AppleIntelligence from '@pxltail/apple-intelligence';
const PORT = 8080;

const ai = AppleIntelligence(PORT);
ai.start(); // Server started at localhost:8080

• If you're using this outside of JS in a terminal, there's also a CLI tool:

apple-intelligence 8080 # Server started at localhost:8080

Both of these start an HTTP server at the port provided.

Usage

The server you started has OpenAI API compatible endpoints. Here's what is and isn't supported.

If you know how to use the OpenAI API, this is a simple drop-in. If not, there are tools like Msty you can use that make interacting with local models very easy.

This server also strips /v1/ prefixes, so if you place it into a system that does that, it will still work.

Example Request

curl -X POST http://localhost:8080/chat/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "afm",
        "messages": [
            { "role": "system", "content": "You are a helpful assistant." },
            { "role": "user", "content": "What is the capital of France?" }
        ],
        "temperature": 0.7,
        "max_completion_tokens": 100,
        "stream": false
    }'

Expected response:

{
    "object": "chat.completion",
    "created": 1720632342342,
    "model": "afm",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "The capital of France is Paris."
            },
            "finish_reason": "stop"
        }
    ]
}

Issues

This project is still in beta, just like the AI itself. If you find any issues with this project, let me know by @ing me on Twitter. (@pxltail)

This is my first project using Swift, so it may be buggy.. but any major errors have been fixed. If you do get an error, let me know how to reproduce it and I'll try my best to fix it.

The code for this project...isn't great. If you have any suggestions or help related to that, also let me know.

PRs are also welcome! https://github.com/PxlTail/Apple-Intelligence/pulls