MackingJAI

Accessing Apple Intelligence and ChatGPT desktop through OpenAI / Ollama API

Overview

MackingJAI mocks the process of using OpenAI API or Ollama API for chat models by using Apple Intelligence models or ChatGPT Mac app. Users can simplify the interaction with Apple Intelligence models or ChatGPT without using an API key.

You can use it for any application that uses OpenAI API or Ollama API to run.

Notes:
To use Apple Intelligence models, you need to have MacOS Tahoe 26

Supported Models

Supported Models:

• Apple Intelligence:

  • apple_local (On device apple intelligence model)
  • apple_cloud (Private Cloud Compute apple intelligence model)

• ChatGPT:

With the release of new GPT 5 models, ChatGPT shortcut doesn't have an option to set the model. It will route the request automatically. Just set it to GPT-5 and it will work!

Installation

• Download and install the DMG file from releases
• Install the shortcut by clicking on Install Shortcut from the menu icon or from here

• Ensure that:
  • For Apple Intelligence models, Apple Intelligence is enabled on macOS Tahoe 26 or later.
  • For ChatGPT models, the ChatGPT Desktop app is installed and running on your Mac.
• In any OpenAI API compatible request, set the API base to http://127.0.0.1:11435/v1/ instead of https://api.openai.com/v1 and set any value for the API key. The API key is not used in this mock.
• If you use the Ollama API, change the base URL from http://127.0.0.1:11434 to http://127.0.0.1:11435. No further changes are required.

Note: You may be asked to give permissions (e.g., in System Settings under Accessibility or Automation) to allow the shortcut to intercept API calls. This is required for it to work correctly.

Usage

Theory, you can use any OpenAI API compatible library to make requests to the mocked API. However there are many limitations discussed below.

Curl:

curl http://127.0.0.1:11435/v1/chat/completions \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer No need for API" \
    -d '{
        "model": "apple_cloud",
    "messages": [
        {
            "role": "system",
            "content": "You are a fact checker, answer with yes or no only"
        },
        {
            "role": "user",
            "content": "Paris is the capiatl of France"
        }
    ]
}'

OpenAI Python

from openai import OpenAI
client = OpenAI(api_key="No need for API", base_url="http://127.0.0.1:11435/v1/")

completion = client.chat.completions.create(
  model="GPT-5",
  messages=[
    {"role": "developer", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Hello!"}
  ]
)

print(completion.choices[0].message)

LangChain 🦜🔗

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model_name="GPT-5",
    openai_api_base="http://127.0.0.1:11435/v1",
    openai_api_key="No key",
)
messages = [
    (
        "system",
        "You are a helpful assistant that translates English to French. Translate the user sentence.",
    ),
    ("human", "The quick brown fox jumps over the lazy dog"),
]
ai_msg = llm.invoke(messages)
print(ai_msg)

Open-webui

Just add a new Ollama API connection with the URL http://host.docker.internal:11435

• If you don't use docker, check the URL of Ollama in Ollama API and replace 11434 with 11435 in the URL

Tip: If you are using reasoning models like o3 or o4-mini, it's better to set a different model for Title and Tag Generation.
Navigate to Settings > Admin Settings > Interface > Set Task Model and choose a non-reasoning model. This will save your quota and speed up the generation process.

Raycast AI

MackingJAI can be used for Raycast AI, just change Ollama Route from 11434 to 11435 in the Raycast AI settings. It will access all available models in your ChatGPT Desktop app.

GitHub Copilot

To use MackingJAI with GitHub Copilot, you need to set the Ollama endpoint to http://localhost:11435 in the GitHub Copilot settings. Then select the models you want to use from the dropdown menu.

Limitations

• Everything is limited by your apple intelligence or chatgpt desktop application and your subscription including available models, rate limits and generation speed.
• There's no way use any other parameters like temperature, top_p, etc.
• You can't send images in this mock.

Todo

• Explore how to integrate conversation history ✅
• Explore how to integrate system prompt if possible ✅
• Create a homebrew cask for easy installation
• Explore similar functionality for Windows users