We make Pinku, the app this guide uses. The idea is the same in every offline AI app: the model is downloaded once and then runs on your device.
Cloud assistants send every message to a server, so they stop working on a plane, in a tunnel or where they are not offered. An offline AI app keeps a language model on the device itself. You need the internet once, to download it; after that every answer is generated on the iPhone, and nothing you type leaves it.
Step by step
- Install Pinku. Get Pinku from the App Store. It runs on iPhone, iPad and Mac with Apple silicon, on devices that support Metal 3. There is no account.
- Download a model while you are online. Do this on Wi-Fi before you need it. The default model, Gemma 3 1B, is under 1 GB. Pinku only lists the models that fit your device's memory.
- Turn on airplane mode and test it. Open Control Center, turn on airplane mode and ask a question. The answer is generated on the device, so it works with no connection at all.
- Pick the right model for the job. Smaller models answer faster; larger ones answer better. The six models up to 1 GB are free; Pinku Pro, a one-time purchase, adds six larger ones up to 4.7 GB.
- Start a chat with Siri, if you like. Pinku has a Siri and Shortcuts action to start a chat, which also works offline.
- Move a chat when you need to. Chats stay on the device that wrote them; there is no sync. Share a conversation as plain text to keep it somewhere else.
What to expect offline
A model of 1 to 5 GB on a phone is clearly weaker than ChatGPT or Claude in the cloud. It is good for drafting, rewriting, summarising text you paste in, explaining things and brainstorming. It cannot search the web, does not read images or documents, has a shorter memory of the conversation, and its speed depends on your device. Pinku checks the memory actually free before loading a model, so a model that does not fit gives a warning instead of a crash.
Pinku is free to download, and the six models up to 1 GB work without paying. Pinku Pro ($9.99 on the US App Store, one time, no subscription) adds the larger models, custom instructions, dictation recognised on the device and read aloud. More on Pinku's page.
Other offline AI apps, including where they beat us: the best offline AI chat apps for iPhone and Mac.
Questions people ask
Can I use AI on my iPhone without internet?
Yes, with an app that runs the model on the device, such as Pinku. You need a connection once to download the model; after that it answers in airplane mode.
Does ChatGPT work offline?
No. ChatGPT runs on OpenAI's servers, so every message needs a connection.
How much space does an offline AI model take?
In Pinku, from 0.34 GB to 4.7 GB depending on the model. The default, Gemma 3 1B, is under 1 GB.
Is an offline AI as good as ChatGPT?
No. A small model on a phone is clearly weaker, has a shorter context and cannot browse the web. Its advantage is that it works anywhere and nothing you type leaves the device.
Are my conversations sent anywhere?
Not with Pinku. There is no account and no server; chats are stored on the device. The network is used only to download models and for App Store purchases.