PREREQUISITES AND HARDWARE

Before we install anything, let's make sure your machine can handle it. Local AI is demanding. Not gaming-PC demanding, but your 2015 MacBook Air isn't going to cut it.

Check Your Chip

Open Terminal and run:

sysctl -n machdep.cpu.brand_string

You need Apple Silicon. That means M1, M2, M3, or M4, any variant. The Pro, Max, and Ultra versions are better because they have more memory bandwidth, but the base chips work fine.

Intel Macs can technically run this but performance is terrible. If you're on Intel, tonight might be more of a watch-and-learn session.

Check Your RAM

sysctl -n hw.memsize | awk '{print $1/1024/1024/1024 " GB"}'

This is the most important number. Local AI models load entirely into RAM. Here's what you can realistically run:

8GB:    Small models only (4B parameters or less)
16GB:   Medium models, one at a time, with tight context
32GB:   Most models comfortably, including the good ones
64GB+:  Multiple models, large context windows, the works

The sweet spot for tonight is 32GB. That gets you access to the best models we'll be demonstrating.

Why RAM Matters More Than You Think

On a traditional PC, your system memory and your graphics card memory are separate pools. A model has to fit in one or the other.

Apple Silicon is different. The CPU and GPU share the same unified memory pool. If your Mac has 64GB of unified memory, the AI model gets access to all 64GB. No artificial split.

This is why a Mac Studio with 64GB of unified memory can run models that would require a $2,000 NVIDIA graphics card on a PC. Apple accidentally built one of the best consumer AI platforms.

Disk Space

You'll need roughly:

Colima + Docker:          ~1GB
OpenWebUI container:      ~2GB
Per AI model:             3GB to 50GB each

Plan for at least 30GB free. More if you want to experiment with multiple models.

Software Prerequisites

We need two things installed before OpenWebUI:

1. Colima (plus the Docker CLI tools)
2. Ollama

If you already have Ollama from a previous session, you're halfway there. Check that it's running:

ollama --version

You want 0.14.0 or higher.

Installing Colima and Docker

We're going to use Colima instead of Docker Desktop. Colima is a free, open source, command-line Docker runtime for Mac. No GUI app, no license restrictions, just terminal commands.

Here's the thing most tutorials skip: Docker on Mac requires three separate packages, and they each do a different job.

1. colima = the Linux VM (Docker needs Linux to run containers)
2. docker = the CLI tool (so you can type docker commands)
3. docker-compose = the orchestration tool (runs multi-container setups)

Without all three, it doesn't work. If you install Colima but skip the docker package, typing docker gives you "command not found."

Install all three with Homebrew:

brew install colima docker docker-compose

Register Docker Compose as a Plugin

Docker Compose installs as a standalone binary, but Docker doesn't know about it by default. You need to register it as a plugin so that "docker compose" (with a space) works:

mkdir -p ~/.docker/cli-plugins
ln -sfn /opt/homebrew/opt/docker-compose/bin/docker-compose \
  ~/.docker/cli-plugins/docker-compose

Without this step, you'll get "docker: 'compose' is not a docker command" when you try to use it.

Starting Colima With Proper Resources

This is the step that trips people up. If you just run colima start with no flags, you get the defaults: 2 CPUs and 2GB of RAM. That's not enough to run anything useful.

Start Colima with proper resources:

colima start --cpu 4 --memory 8 --disk 60

What those flags do:

--cpu 4           Give the VM 4 CPU cores
--memory 8        Give it 8GB of RAM
--disk 60         60GB of disk space

Adjust the numbers based on your Mac. Don't give Colima more than about half your total RAM. A 32GB Mac should be fine with 8GB for Colima, leaving plenty for Ollama and the rest of macOS.

If you're on real Apple Silicon hardware (not a VM), you can make Colima faster by using Apple's native virtualization:

colima start --cpu 4 --memory 8 --disk 60 \
  --vm-type vz --vz-rosetta --mount-type virtiofs

The vz flags use Apple's Virtualization.framework instead of QEMU. Noticeably faster, but only works on real hardware. If you're running macOS in Parallels or UTM, stick with the basic command.

These settings are saved automatically. Next time you run colima start with no flags, it remembers the configuration.

Verify everything is working:

docker ps

You should see an empty container list. If you see "command not found," you forgot to install the docker package. If you see "cannot connect to Docker daemon," Colima isn't running.

Auto-Start on Login

Without this, after every reboot you'll open Terminal, type docker, and get "cannot connect." You have to manually start Colima first every single time. Fix that with:

brew services start colima

Now Colima starts automatically when you log in. Docker commands just work from the moment you open Terminal.

To stop Colima manually when you need to:

colima stop

To start it again:

colima start

Installing Ollama

If you don't have Ollama yet:

brew install --cask ollama

Or download from ollama.com. After installing, open the Ollama application. It runs as a menu bar icon and serves an API on port 11434. You won't interact with it directly most of the time. It just needs to be running in the background.

Verify:

ollama --version

Make sure it shows 0.14.0 or higher. If you're on an older version:

brew upgrade --cask ollama

Quick Checklist

Before moving on, confirm:

[ ] Apple Silicon Mac with 16GB+ RAM
[ ] brew install colima docker docker-compose (all three!)
[ ] Docker Compose registered as plugin (ln -sfn command)
[ ] Colima started with proper resources (--cpu 4 --memory 8)
[ ] docker ps works (shows empty container list)
[ ] brew services start colima (auto-start on login)
[ ] Ollama installed and running (0.14.0+)
[ ] At least 30GB free disk space

All good? Next we'll install OpenWebUI itself.