Locally AI

Locally AI is a native iPhone, iPad and Mac app that runs open-source models such as Llama, Gemma, Qwen and DeepSeek fully offline on Apple Silicon, with no login and no cloud processing.

Locally AI is a native Apple app that runs open-source language and vision models directly on an iPhone, iPad or Mac. Once a model is downloaded, chat works without an internet connection, without an account and without any cloud processing. The developer states that data stays on the device, which is the central reason to choose it over hosted assistants. The app is aimed at people who want a private assistant without sending prompts to a server: privacy-minded professionals, students working offline, and hobbyists curious about on-device inference. Setup is short. A user installs the app from the App Store or Mac App Store, picks a model to download, and starts chatting. Supported families include Llama, Gemma, SmolLM, DeepSeek, Qwen, Granite, Cogito and Liquid AI's LFM, with named versions such as Llama 3.2, Gemma 3n, Qwen 3 and DeepSeek R1. Performance rests on Apple Silicon and MLX, Apple's machine learning framework built around unified memory. The app also uses Apple Foundation Models where the operating system provides them. Beyond text chat it offers an on-device voice mode, image understanding through vision models, and an editable system prompt for shaping responses. System hooks include a Siri phrase, Control Center, Lock Screen and Action Button access, and Apple Shortcuts support for automation. Larger models are meant for iPad and Mac, where more memory is available, while phones suit smaller ones. The site claims results comparable to GPT-4 and GPT-4o-mini on that hardware, though an asterisk qualifies the claim. Among alternatives, it sits in the same space as desktop runners and other mobile local-LLM apps, but it is restricted to Apple devices. The homepage also announces that Locally AI is joining LM Studio, so the product's future direction may change.