is there an ai that can be used offline?

If you want a clear offline promise

LM Studio, now called LM Studio Bionic, downloads open models and runs them on your computer. Its docs say chatting with models, chatting with documents and running a local server all work without the internet, and that nothing you enter leaves your device. The cost is the hardware: LM Studio recommends at least 16GB of RAM and, on Windows, at least 4GB of dedicated graphics memory. On a Mac it needs Apple Silicon and macOS 14 or newer. Its free plan at $0 runs local models, according to its pricing page, and it runs on Mac, Windows and Linux.

If you want open source with no account

Jan is an open-source chat app its makers call a replacement for ChatGPT and Claude, running on your own hardware. It is Apache 2.0 licensed, says its desktop apps have full offline support, and its quickstart needs no account. Jan's homepage describes it as free and open source. The catch is that Jan can also plug in online models such as ChatGPT and Claude, and those are not offline, so you need to keep track of which model you are talking to. Its memory feature is marked coming soon, so do not count on that yet.

If you want to plug local models into coding tools

Ollama is a tool for downloading and running open models on your computer. It says local models are always free, and that it can launch Claude Code, Codex and other agents with one command. What you give up is focus: its homepage leads with paid cloud models, which are not offline, so the free local side is easy to miss. The Mac app needs macOS 14 or later, and there are versions for Windows and Linux too.

If your computer is older or has no graphics card

GPT4All is Nomic's local AI chatbot for desktops and laptops. Its README says no API calls or GPUs are required, and the Windows and Linux builds run on an Intel Core i3 2nd Gen or AMD Bulldozer or better. It also has LocalDocs, for chatting with your own files. Nomic's GPT4All page offers downloads with no price shown. What you give up is certainty about the future. The latest GitHub release, v3.10.0, is from 25 February 2025, and Nomic's site now centres on its business platform, so check it is still maintained before you rely on it.

If you want it on your phone

Google AI Edge Gallery is the phone option here. It runs open models such as Gemma 4 on an Android phone or iPhone, and the listings say all model inference happens on the device and no internet is required, with chat, image questions and audio transcription. The UK Google Play listing shows Install with no price and no in-app purchases, and the UK App Store lists it as free. How well it runs depends on your phone's processor and graphics hardware.

When paying is worth it

Never, for offline use. Ollama and LM Studio both sell cloud plans now: Pro at Ollama is $20 per month, or $200 per year billed annually, and Bionic+ at LM Studio is $20 USD per month, with Pro at $100 USD per month. Those cloud models run on their servers, so they need the internet. Ollama's paid plans are built around usage credits for its cloud models, and its $0 plan includes running models locally. LM Studio's $0 plan runs local models too. If your goal is offline use, the paid tiers solve a different problem.

The first point is that you need the internet before you can go offline. LM Studio's docs say searching for and downloading models, downloading runtimes and checking for updates all need a connection. Download the models you want while you are online.

Mixed apps can send some requests online. Jan can plug in online models such as ChatGPT, Claude and Gemini, and only its local models stay on your machine. The same applies to the cloud plans at Ollama and LM Studio. Ollama says nothing you run locally ever leaves your machine, and LM Studio says the same for chats with downloaded models, but cloud features in the same apps are a different matter. If privacy is the reason you want offline AI, check which model is selected.

Apple Intelligence is not fully offline either: Apple says it can draw on larger server-based models through Private Cloud Compute. And Google says AI Edge Gallery is in active development, with performance that depends on your phone's hardware, so expect it to change.