AI Without Internet: What Works Offline on Your Phone

Yes, you can use AI on your phone with no internet connection. Two kinds of AI work offline: features built into iOS and Android that run on the phone’s own chip, and local AI apps that download an open language model once and then answer everything on the device. Cloud chatbots like ChatGPT, Gemini and Claude do not work offline at all.

The catch is that “AI” on a phone covers a dozen different features, and some of them quietly depend on a server. Here’s what actually keeps working in airplane mode.

What AI works offline on an iPhone? #

Apple runs a lot of its machine learning on the device, but not everything branded Apple Intelligence stays local.

• Apple Intelligence Writing Tools. On an iPhone 15 Pro or newer, rewriting, proofreading and summarizing text use Apple’s roughly 3-billion-parameter on-device model for most requests. Harder requests can go to Apple’s Private Cloud Compute servers, which need a connection.
• Translate. The Translate app can work offline once you download the languages you need.
• Dictation. On recent iPhones, dictation in many languages is processed on the device.
• Photo search and Live Text. Finding “dog” or “receipt” in your photo library and copying text out of an image both run locally.

What doesn’t work offline: Siri requests that need the web, anything handed to ChatGPT through Apple’s integration, and the larger Apple Intelligence features that run on Private Cloud Compute. We break down which requests go where in how private Apple Intelligence really is.

What AI works offline on Android? #

Google’s on-device model is Gemini Nano. It runs inside a system service called AICore, and Google’s own documentation says it works “without needing a network connection or sending data to the cloud.” Apps can use it for summarizing, proofreading, rewriting, describing images and transcribing speech.

The limit is hardware. Gemini Nano only ships on specific phones, including the Pixel 9, 10 and 11 series, the Galaxy S25 and S26 series, recent Galaxy Z foldables and a list of flagships from OnePlus, Xiaomi, vivo, OPPO and others. A mid-range phone from 2022 won’t have it.

Beyond Gemini Nano, a few long-standing Android tools work offline:

• Google Translate with downloaded language packs, including camera translation for downloaded languages.
• Recorder on Pixel phones, which transcribes recordings on the device.
• Live Transcribe and Live Caption for real-time captions, depending on the phone and language.

What still needs a connection? #

This table covers what people usually ask about.

Local models also can’t tell you about anything that happened after their training data was collected. We cover that limit, and how to work around it, in can offline AI know current events.

How do you get a full AI chatbot that works offline? #

Built-in features handle small jobs like fixing a sentence or translating a sign. If you want a real conversation with an AI, the kind of back-and-forth you’d have with ChatGPT, you need a local LLM app. The setup takes about ten minutes.

1. Install a local AI app. It should say plainly that models run on the device. Personal LLM is free on iOS and Android, needs no account, and runs open models such as Qwen 3.5, Gemma 4, Ministral 3 and GLM 4.6V Flash on the phone’s own chip.
2. Connect to Wi-Fi. Model files are big, from about 0.8 GB to over 6 GB.
3. Pick a model that fits your phone. Personal LLM reads your phone’s RAM and shows a “Fits your device” badge on each model before you download. Qwen 3.5 4B (2.74 GB) is its recommended first download for most phones.
4. Download it once. Downloads can be paused and resumed if the Wi-Fi drops.
5. Test it in airplane mode. Turn on airplane mode, open a new chat and ask something. If it answers, it’s truly local.

After that, the app doesn’t need the internet to chat, read a PDF you attach, or describe a photo if you add the model’s image-support file. The only things in Personal LLM that use the network are model downloads and the ads in the free version, and ads simply don’t load when you’re offline.

How good is offline AI compared to ChatGPT? #

Be realistic about this. A 4-billion-parameter model on your phone is much smaller than the models behind ChatGPT or Gemini, and it shows on hard reasoning, niche facts and long documents. Small models also make things up more often, so check facts that matter.

Where local models do well:

• Rewriting, summarizing and proofreading text you give them
• Explaining a concept you half-remember
• Brainstorming, drafting emails, planning a trip
• Answering questions about a document or photo you attach
• Practicing a language

Where they fall short:

• Anything recent, since there’s no web access
• Long multi-step math without checking (thinking mode helps)
• Obscure facts, citations and exact quotes
• Very long documents, since phone RAM limits the context window

If you mostly need quick, private help with your own text, a local model covers it. If you need the strongest reasoning available, a cloud chatbot will beat it, and you’re trading away privacy and offline use for that.

Who benefits most from offline AI? #

• Travelers. Flights, subway commutes, roaming abroad without a data plan.
• People with patchy coverage. Rural areas, basements, hospitals, big concrete buildings.
• Anyone handling private text. Medical notes, contracts, journal entries, work drafts you can’t paste into a cloud service.
• People who hit data caps. After the one-time download, chatting uses no data.

Offline AI is a poor fit if you want live information, image generation, or answers at the quality of the biggest cloud models every time.

Tips before you go offline #

• Download before you leave. A 3 GB model over hotel Wi-Fi is no fun.
• Add image support if you’ll want it. In Personal LLM each model’s image-support file is a separate 195 MB to 1 GB download, and you’ll want it for menus and signs abroad.
• Check your storage. Leave room for the model plus a few gigabytes for the phone itself. See how much storage AI models need.
• Carry a smaller backup model. If battery gets low, a smaller model answers faster and uses less power.

Frequently asked questions #

Does ChatGPT work offline? #

No. ChatGPT runs on OpenAI’s servers, so the app needs a connection for every message. Your past chats may still be visible offline, but you can’t get new answers. For offline chat you need an app that runs a model on your phone.

Can I use AI on a plane without Wi-Fi? #

Yes, if you downloaded a local model before boarding. Local AI apps work in airplane mode because the model and the processing are both on the phone. Built-in features like downloaded Translate languages also work.

Does offline AI use mobile data? #

Not for chatting. The model download uses data once, so do it on Wi-Fi. After that, answers are generated on the phone. In Personal LLM, ads in the free version load only when you’re online.

What phone do I need to run AI offline? #

For local LLM apps, less than you might think. Personal LLM runs on iOS 15.1 or later and Android 7 or later, and its smallest model needs about 2 GB of RAM. Apple Intelligence needs an iPhone 15 Pro or newer, and Gemini Nano only runs on recent flagships. We cover older hardware in running AI on an old phone.

Is offline AI more private than cloud AI? #

Yes, by design. If the model runs on your phone and the app has no server, your messages have nowhere to go. Check that an app is truly local by testing it in airplane mode and reading its privacy label.