The advent of artificial intelligence (AI) has opened vast windows of opportunity for automation and to assist humans with various tasks. However, the nature of the most popular cloud AI providers is not private.
Storing your data on another person’s computer (which means anything cloud related) means that you are one decision away from having your data misused.
Distillation and other recent AI training techniques have resulted in new AI models that are a fraction of the size of frontier models, and can be run on your phone without using your mobile data nor sending your messages to a server.
This is a game changer for anyone who wants to keep their intellectual property (IP) private when bouncing ideas or plans off AI models, and anyone else who doesn’t want their privacy invaded.
Modern small AI models have minuscule system resource and energy usage compared to the most popular frontier models. This is what enables them to run on mobile phones, laptops and desktops. This also translates to a significantly lower cost.
A Few Examples Of Mobile Phone-Compatible Models
- Gemma 4 E4B.
- Gemma 4 E2B (smaller and works on more phones).
- Microsoft Phi-4-mini (designed for mathematics).
- DeepSeek R1-Distill-Qwen-7B.
You can easily get started with offline, private AI models on your phone using the Off Grid AI app for iOS, or on Android.
The Off Grid AI app can download and run some small AI models on your phone automatically. However, it can also connect to an LM Studio or similar installation on your computer. This means you can use your phone to chat with much larger models offline — but without having to store them on your phone.
LM Studio could be considered the desktop equivalent of Off Grid AI, as it offers the same functionality with the ability to install much larger models (models as large as your computer can accommodate). LM Studio makes it easy to install AI models on Linux, MacOS and Windows.
Image credit: Pexels.
