How to use local AI from your phone

Cloud chatbots see everything you type. A local LLM keeps your chats on your own hardware. This guide shows three ways to use local AI from an Android phone or iPhone, and what each one takes.

Updated November 1, 2026 · 7 min read

Why use a local LLM at all?

• Privacy. Your questions, files and photos never leave your devices. Nothing is stored on someone else’s server or used to train their next model.
• Your data stays in your country. Most cloud AI services process what you type on servers abroad. For a business handling customer, patient or employee data, that transfer can fall under laws like GDPR and KVKK. A local model keeps the data on your own premises.
• No subscriptions or limits. The model runs on hardware you already own.
• Works without the internet. Once a model is downloaded, it keeps working with no connection.

The catch is size. Good models need lots of memory, and phones have little. That’s why the three options below differ mostly in where the model runs.

Option 1: run a small model on the phone

Several apps download a model straight to your phone and run it there. It’s the most private option and works on a plane. But a phone can only hold small models (around 1 to 4 billion parameters), so answers are simpler, long chats run out of room sooner, and the phone gets warm and drains battery.

Good for: quick questions with no internet. Not great for: writing, coding, long documents.

Option 2: connect your phone to Ollama on your computer

Ollama is a free program that runs open models on a Mac or PC. Your computer has far more memory than your phone, so it can run much bigger, smarter models. To reach it from your phone yourself:

1. Install Ollama on your computer and download a model, for example: ollama pull llama3.2
2. Let other devices reach it. By default Ollama only answers the computer it runs on. On a Mac, turn on Ollama’s Expose Ollama to the network setting, or run launchctl setenv OLLAMA_HOST 0.0.0.0 and restart Ollama. On Windows, add a user environment variable OLLAMA_HOST with the value 0.0.0.0, then quit and reopen Ollama.
3. Find your computer’s local IP address (on a Mac: System Settings → Wi-Fi → Details; on Windows: Settings → Network & internet → your network’s properties, under IPv4 address), for example 192.168.1.20.
4. Install an Ollama client app on your phone (there are several for Android and iPhone) and enter http://192.168.1.20:11434 as the server.

It works, but there are trade-offs to know about:

• No password. Once Ollama listens on your network, anyone on the same Wi-Fi can use it.
• Addresses change. When your router gives the computer a new IP address, every phone stops working until you update it.
• One setup per phone, per person. There’s no simple way to see who is using it or to limit a family member to certain models.
• You keep it running. If Ollama isn’t running, or the model is too big for your memory, the phone just shows an error.

Option 3: use PetekMind

PetekMind does option 2 for you, without the trade-offs. PetekMind Hub runs on your computer (Mac today, Windows coming soon) and the PetekMind app runs on Android and iPhone. If you were looking for an Ollama Android app or a way to use your computer’s AI from your iPhone, this is it, with the setup done for you:

1. Install PetekMind Hub on your computer. It installs and starts the AI engine (Ollama) for you, keeps it running in the background, and suggests the best model for your computer’s memory.
2. Pair your phone. The app finds your computer on the Wi-Fi. Scan a QR code or match a 4-digit code, and you’re in. No IP addresses or ports.
3. Chat. Messages go from your phone to your computer and back. Chats are saved on your phone.

Only phones you approve can use it, and you can pause a device or limit a person to certain models at any time. Away from your Wi-Fi, the app can still run a small model on the phone itself (option 1), so you’re never stuck.

Which model fits your computer?

The main limit is memory. A model needs roughly its download size in memory, plus room for the conversation and for your other apps. On a Mac that’s the regular memory (RAM); on a Windows PC with a graphics card, it’s mostly the card’s memory (VRAM). As a rough guide:

PetekMind Hub does this math for you: it marks the biggest all-round model that fits comfortably in your computer’s memory as Best fit, and won’t start a download that would fill up your disk.

Quick comparison