How to Run an LLM Locally on Your Mac: Ollama vs LM Studio

What you need to run an LLM locally on a Mac

Any recent Mac can run a small model. How well it runs comes down to two things: Apple silicon versus Intel, and how much memory you have. Here’s what each app asks for, from their own docs:

To see your chip and memory, choose Apple menu › System Settings › General › About and read the Chip (or Processor) and Memory lines.

Note: Rule of thumb, not a spec: 8 GB runs 3–4B models (2–3.4 GB downloads), 16 GB runs 7–12B models, and 24 GB or more opens up the 14–27B range. You need room for the model, the text it’s working on, and macOS itself.

How to install Ollama on a Mac

1. Download Ollama for macOS from ollama.com and open the .dmg.
2. Drag Ollama into your Applications folder, then open it.
3. When it asks, let it create a link in /usr/local/bin. That puts the ollama command in Terminal.
4. Open Terminal and run your first model:

ollama run llama3.2

The first run downloads Llama 3.2 3B (about 2.0 GB), then starts a chat right in Terminal. Type a question, press Return, and type /bye to leave.

Rather install from Terminal? Ollama’s download page also gives a one-line install script:

curl -fsSL https://ollama.com/install.sh | sh

With Homebrew, brew install ollama installs the command-line version, and brew services start ollama starts its server now and every time you log in.

Tip: Not a Terminal person? The Ollama app has its own chat window, where you can download and chat with models without typing a command.

Everyday Ollama commands

• ollama pull qwen3:4b downloads a model without starting a chat.
• ollama ls lists the models you’ve downloaded.
• ollama ps shows which models are loaded in memory right now.
• ollama stop qwen3:4b unloads a model from memory.
• ollama rm qwen3:4b deletes it from your disk.

Where Ollama keeps models, and its local address

• Models live in ~/.ollama/models in your home folder. Logs are in ~/.ollama/logs.
• The server listens on 127.0.0.1:11434, so only apps on your Mac can reach it.
• Apps that speak the OpenAI API can use http://localhost:11434/v1.

Short on disk space? Set OLLAMA_MODELS to keep models on another drive. For the Mac app, Ollama’s docs say to set environment variables with launchctl, then quit and reopen Ollama:

launchctl setenv OLLAMA_MODELS "/Volumes/External/ollama-models"

How to install LM Studio on a Mac

1. Download LM Studio for Mac from lmstudio.ai and install it like any other app. It needs an Apple silicon Mac with macOS 14 or later.
2. Open the Discover tab. Pick one of the suggested models, or search by name (for example, “Qwen3”), and download it.
3. Go to the Chat tab, open the model loader, and choose the model you downloaded.
4. Start chatting. Everything runs on your Mac.

Turn on LM Studio’s local server

To let other apps use the model, open the Developer tab and switch on Start server. Or run this in Terminal (the lms command comes with LM Studio; open the app once first):

lms server start

Apps can then reach it at http://localhost:1234/v1, an OpenAI-compatible address. Since July 2025, LM Studio is free to use at work too, with no form to fill in.

Ollama vs LM Studio: which should you use?

Pick LM Studio if you have Apple silicon and want to browse, compare, and chat with models in a proper app. Pick Ollama if you have an Intel Mac, like the command line, or want a server that other apps can use with no extra steps. You can install both, but each downloads its own copy of a model, so keep an eye on disk space.

Which model should you start with?

• 8 GB Mac: llama3.2 (2.0 GB) or qwen3:4b (2.5 GB).
• 16 GB: qwen3:8b (5.2 GB) or gemma3:12b (8.2 GB).
• 24 GB and up: qwen3:14b (9.3 GB), or gemma3:27b (17 GB) if you have 32 GB.

Sizes are Ollama’s downloads; in LM Studio, search for the same names in Discover. For picks aimed at writing, with languages and licenses side by side, see the best local LLM for writing on a Mac.

How to free up memory and disk space

• Memory: Ollama keeps a model loaded for 5 minutes after you last use it, then unloads it. Run ollama stop with the model’s name to unload it right away, or set OLLAMA_KEEP_ALIVE to change the timer.
• Disk: ollama ls lists your models and their sizes, and ollama rm deletes one. In LM Studio, the My Models tab shows what you’ve downloaded.
• Uninstalling Ollama: quit it, then delete /Applications/Ollama.app, /usr/local/bin/ollama, and ~/.ollama (that last folder holds your models). Ollama’s Mac docs list a few more support files in ~/Library for a full cleanup.

Use your local model in every Mac app

Chatting in a window is a start. The bigger win is using the model where you actually write. Both servers speak the OpenAI API, so any app that accepts an OpenAI-compatible address can use them: enter http://localhost:11434/v1 for Ollama or http://localhost:1234/v1 for LM Studio. Ollama ignores the API key, so if an app insists on one, any placeholder works.

In MacWriteBuddy: Choose Ollama or LM Studio in Settings › Providers and pick a model. Then select text in Mail, Slack, Notes, or any app and press ⌃⌥G to fix spelling and grammar, free, with your text staying on your own hardware. Rewrites, Summarize, and Translate come with Pro, a one-time payment of $39 (usually $79).

Don’t want to run a server at all? MacWriteBuddy can download a small model itself and work as an offline grammar checker. Or use a cloud model with your own API key when your Mac is short on memory. It’s free to download.

Troubleshooting a local LLM on a Mac

• It’s very slow on an Intel Mac. That’s expected: Ollama runs on the CPU only there. Stick to the smallest models (llama3.2:1b is a 1.3 GB download), or run Ollama on a faster computer and connect to it over your network. Ollama’s FAQ explains how with OLLAMA_HOST; only do that on a network you trust. More ideas in what works on an Intel Mac.
• Your whole Mac slows down. The model may be too big for your memory. Open Activity Monitor and click Memory: if Swap Used keeps climbing, switch to a smaller model or close other apps.
• Long documents get cut off or forgotten. Ollama’s docs say its default context is 4K tokens on machines with less than 24 GB of GPU memory. Raise it with the context-length slider in Ollama’s settings, knowing that more context needs more memory.
• “ollama: command not found”. Open the Ollama app and accept its offer to link the command into /usr/local/bin, then open a new Terminal window.
• Another app can’t connect. Check that Ollama is open or LM Studio’s server is switched on, and that the address ends in /v1.

Running a model locally is also the simplest answer to “where does my writing go?” If that’s why you’re here, see whether Grammarly is safe for confidential work.