Short answer: partly. Apple Intelligence works offline for the tasks its on-device model can handle. Anything that needs a bigger model is sent to Apple's Private Cloud Compute servers, and anything you hand to ChatGPT goes to OpenAI. Both of those need an internet connection. You don't choose where a request runs; the system decides.
This guide explains which parts work without internet, why the rest doesn't, and what to use when you need AI that works fully offline on a Mac, iPhone, or iPad.
How Apple Intelligence decides where a request runs
Apple Intelligence uses three tiers:
| Tier | Where it runs | Works offline? |
|---|---|---|
| On-device model | Your iPhone, iPad, or Mac | Yes |
| Private Cloud Compute | Apple's servers, for requests too large for the device | No |
| ChatGPT integration | OpenAI's servers, only when you approve it | No |
The on-device model is small enough to run on your phone's chip, so it handles short, focused jobs well. When a request is longer or more complex, Apple Intelligence escalates it to Private Cloud Compute. If you're offline at that moment, the request fails or the feature says it needs a connection.
What usually works without internet
- Many Writing Tools tasks on short text, such as proofreading and rewriting a paragraph.
- Notification and message summaries on the device.
- Image Playground and Genmoji, which generate images on the device.
- Priority messages and smart replies in Mail and Messages.
What needs the internet
- Longer or harder requests that Apple Intelligence routes to Private Cloud Compute, such as rewriting or summarizing long documents.
- Anything sent to ChatGPT, including open-ended questions Siri passes along.
- Questions about the world that need current information from the web.
Apple adds and moves features between tiers with each OS release, so a task that needed the cloud last year may run locally today, and vice versa. The rule stays the same: if the on-device model can't handle it, it needs a connection.
Is Private Cloud Compute private?
Apple built Private Cloud Compute so that your data is used only to answer that request, isn't kept afterwards, and can't be accessed by Apple staff. Apple also publishes the server software images so security researchers can check those claims. That's a much stronger design than a typical cloud AI service.
It is still a remote server, though. Your text leaves the device for that request, and you need a connection. For people bound by confidentiality rules (lawyers, clinicians, finance teams) or anyone working on a plane, in a secure facility, or with patchy signal, "private cloud" and "offline" are not the same thing.
What Apple Intelligence doesn't do: chat with your own documents
Apple Intelligence can summarize the text in front of you, but it doesn't let you build a library of your own PDFs and ask questions about them, with answers that cite the page they came from. That workflow, called retrieval-augmented generation (RAG), needs an app built for it.
How to get fully offline AI on Mac, iPhone, and iPad
If you need AI that never needs a connection, use an app that runs a full open-source language model on your device. Kynora AI is built for this:
- Download a model once from the in-app catalog (Llama, Qwen, Gemma, Mistral, and more). After that, everything runs on your device's chip.
- Private Chat works in airplane mode, with no account and no telemetry.
- Ask Docs indexes your PDFs, Word, PowerPoint, Excel, and text files on the device and answers with page-level citations.
- Voice mode and read-aloud use Apple's on-device speech, so they work offline too.
The trade-off is model size. A model that fits on a phone knows less about the world than the largest cloud models. For questions about your own documents, that matters much less, because the answer comes from the text Kynora AI retrieves from your files. See on-device AI vs cloud AI for the full comparison, and which models run on which devices.
The bottom line
Apple Intelligence is a hybrid: some of it works offline, and you can't make all of it do so. If offline or zero-upload is a requirement rather than a nice-to-have, run a local model. Kynora AI offers a 7-day free trial on Mac, iPhone, and iPad.
Frequently asked questions
Does Apple Intelligence work without internet?
Partly. Features that Apple's on-device model can handle, such as many Writing Tools tasks, notification summaries, and Genmoji, can run without a connection. Requests that need a larger model are sent to Apple's Private Cloud Compute, and ChatGPT requests are sent to OpenAI, so both need the internet.
Does Apple Intelligence on Mac work offline?
The same rules apply on Mac as on iPhone and iPad: on-device tasks work offline, while anything routed to Private Cloud Compute or ChatGPT needs a connection. You can't force every request to stay on the device.
Can I chat with an AI on my iPhone with no internet?
Yes, with an app that runs a full language model on the device. Kynora AI downloads an open-source model once and then runs chat and document questions entirely on your iPhone, iPad, or Mac, including in airplane mode.
Is Private Cloud Compute private?
Apple designed Private Cloud Compute so that data is used only to answer the request and isn't stored or accessible to Apple, and it publishes server software for independent inspection. It is still a remote server, though, so it needs the internet and your data does leave the device for that request.