Michael Limberger

Installing Ollama

Ollama is the local model server. It downloads models, loads them into memory, runs inference on Apple Silicon via Metal, and exposes an API that tools like OpenClaw can call. Think of it as the engine. OpenClaw is the agent layer on top.

Two ways to install

Option A: direct download (recommended). Go to https://ollama.com/download/mac, download the DMG, drag the app to Applications, and launch it. On first run it installs the CLI at /usr/local/bin/ollama.

Option B: Homebrew.

Show me

brew install ollama

That puts the binary at /opt/homebrew/bin/ollama on Apple Silicon. Same result, different path. Direct download is simpler if you do not already use Homebrew.

Verify the install

Open a terminal and run:

Show me

ollama --version

You should see a version number. If you get "command not found", launch the Ollama app once, then open a new terminal.

Show me

ollama list

An empty table is normal before any pulls. A connection error usually means the Ollama app is not running (look for the menu bar icon).

Confirm Metal

Ollama must run as native Apple Silicon to use Metal. Under Rosetta 2, Metal is disabled and inference is painfully slow.

Show me

file $(which ollama)

You want arm64 in the output. If you see x86_64, reinstall from the official Mac download.

Also check the path:

Show me

which ollama

Expected:

/usr/local/bin/ollama (direct install)
        /opt/homebrew/bin/ollama (Homebrew)

An unexpected path may mean a stale install.

+----------------------------------------------------------+
|  NEVER run Ollama under Rosetta 2.                       |
|  It disables Metal entirely. No GPU acceleration.        |
|  Verify with: file $(which ollama)                       |
|  Must show arm64.                                        |
+----------------------------------------------------------+

Do not Docker this

Do not run Ollama in Docker on macOS for this tutorial. Docker sits in a Linux VM without Apple Silicon GPU access, so the model falls back to CPU at roughly 1 to 2 tokens per second instead of tens of tokens per second. Run Ollama natively for Metal.

Start the server

DMG install: launch the app. The llama icon appears in the menu bar and the server starts automatically.

Homebrew, manual start:

Show me

ollama serve

That runs in the foreground with logs. Use another terminal tab for commands.

Verify:

Show me

curl http://localhost:11434/

You should see that Ollama is running. Then you are ready to pull a model.