Locally AI – Local AI Chat

Locally AI is a free Apple app that runs AI language and vision models directly on your iPhone, iPad, or Mac, so your conversations stay on your own hardware. It can use the Apple Foundation model already built into modern iPhones, let you download additional open-weight models, and even hold spoken conversations once you install a voice package. Originally an independent app, it joined LM Studio in April 2026 and now doubles as a mobile front end for the models running on your desktop.

What It Is

Locally AI sits in the local inference category: software that runs AI models on your device instead of sending requests to a cloud server. Out of the box it can talk to the Apple Foundation model — the on-device model the app describes as the same one that powers Apple Intelligence — with nothing to download. From the in-app model manager you can pull in other open-weight models such as Mistral's Ministral, Alibaba's Qwen, and Liquid AI's vision-capable LFM, several of which read images as well as text. The model manager notes that downloadable models are provided by Hugging Face, and it groups them into standard, legacy, and experimental tiers. No account, login, or internet connection is required for on-device chat.

Talking To It, Literally

Beyond typing, Locally AI supports voice. After installing a voice package — the lightweight Kokoro text-to-speech model from Hexgrad — you can hold a spoken back-and-forth with the model, and a built-in Siri Shortcut lets you start a query hands-free by saying "Hey Locally AI." A personalization screen lets you shape how the assistant responds, and conversation history lives on the device, where you can wipe it whenever you want.

LM Link: Borrowing Your Desktop's Muscle

The newest feature, LM Link, connects the app to the full-size models running in LM Studio on your computer. The in-app prompt explains that once Locally has joined LM Studio, you can reach your desktop models on the go from an iPhone or iPad. LM Studio describes the connection as end-to-end encrypted, with chats kept only on your devices — so a phone can either run a small model on its own silicon or tap a much larger one at home without routing anything through a cloud relay.

From Indie App to LM Studio's Mobile Surface

The app began as a standalone project before being folded into LM Studio. The LM Studio blog announced in April 2026 that the Locally AI apps and their creator were joining the team to "double down on Apple platforms," and LM Link shipped alongside an LM Studio desktop update in June 2026. LM Studio is the desktop application many developers already use to run open-weight models on their own machines.

On-Device Privacy by Design

Privacy is the throughline. Because inference happens on the device or over your own encrypted link, the App Store listing states that the developer collects no data from the app. There is no sign-up wall, the app works offline, and you stay in control of both your data and which model you run.