The Rundown

In this guide, you will learn how to download a free AI model to your iPhone and bind it to your phone's Action Button like Siri.

The model runs locally, so after setup you can use it without an internet connection and without sending your private prompts out to a cloud service. It is not going to replace ChatGPT or Claude, but it is a very good way to dip your toes into running local, open source models yourself.

Who This Is Useful For

• iPhone users with an Action Button who want a better assistant shortcut than Siri
• Privacy-focused users who want more AI tasks to run on-device
• People curious about local AI who want an easy first model to test
• Anyone who wants a quick voice assistant for explaining concepts, translating text, or working through small tasks

What You Will Build

You will turn your iPhone's Action Button into a shortcut for an AI that runs on your phone without need for internet or cellular connection.

The initial setup still needs internet because you have to download the model files.

What You Need

• An iPhone with an Action Button (iPhone 15 or newer)
• Locally AI app
• 3+ gigabytes of storage space

Step 1 Download Locally AI

Open the App Store and download Locally AI.

Once it installs, open the app and continue through the welcome screen.

Step 2 Choose Your First Model

When the app asks you to choose a model, we recommend starting with the latest Gemma model available in the app.

Pro tip: Do the downloads on Wi-Fi before you need the assistant offline. Most models are several gigabytes.

Gemma is Google's open-source model family and it is a strong first pick for this setup. It is good enough for normal assistant tasks, and it gives you a clean baseline before you start testing bigger models.

Pro tip: Model names usually include sizes like 4B, 6B, or 8B. That number is roughly the model's parameter count. Bigger usually means more capable, but it also means a bigger download and slower performance on your phone.

Step 3 Download the Model

Tap download and keep the app open while it finishes.

This part may take a few minutes. That is normal. You are downloading the actual model file onto your phone, so it can be a few gigabytes.

Once the model is installed, try one quick prompt before you connect it to the Action Button.

Give me a quick plan for the three most important things I should do before a 30-minute client call.

This test gives you a feel for speed and answer quality. If it responds well enough for quick assistant tasks, you are ready to bind it to your phone.

Step 4 Search for Action Button

Swipe down on your home screen for Action Button.

Open the Action Button settings. Swipe through the options until you see Shortcut.

Step 5 Choose Locally AI Voice Mode

Tap whatever is currently set as the shortcut, search for Locally AI.

Select Voice Mode.

Your Action Button should now be set to Locally AI Voice Mode.

Press the Action Button and wait for the chime. The first time you open Voice Mode, Locally AI may ask you to download a speech-to-text model. Download it.

That speech-to-text model is what lets the app handle voice input locally. It is a separate download from the AI model itself.

Pro tip: If Voice Mode does not launch right away, go back into the Action Button settings and confirm the selected shortcut says Locally AI Voice Mode, not New Chat.

Step 6 Test Your Local Assistant

Now press the Action Button and ask a question out loud.

Try something practical:

Explain the difference between local AI and cloud AI in simple terms.

Or:

Translate "I will send the document this afternoon" into Spanish.

Or:

Give me three ways to prepare for a quick client call.

In our test, the local model was useful, but it was not perfect. One smaller model missed a couple of basic factual questions. Local models are getting better fast, but they can still be less reliable than the huge cloud models you use in ChatGPT or Claude.

For everyday assistant tasks, that tradeoff can still be worth it. Explaining concepts, translation, quick planning, and simple rewrites are all good fits.

Pro tip: After the model and speech-to-text model are downloaded, turn on airplane mode and ask another question. If it still responds, you have proved the important part: the assistant is running locally on your phone.

Going Further

Once the newest Gemma model is working, download a bigger model and run the same prompt through both.

Compare:

• how fast each model responds
• how much storage each model takes
• how accurate the answers are
• which one sounds better out loud

The bigger model may reason better, but the smaller one may be fast enough for everyday voice assistant tasks. That is the real point of testing local models on your phone. You are not just picking the "best" model in theory. You are finding the model that is useful enough, fast enough, and private enough for the tasks you actually want a phone assistant to hand