How to Install Ollama on macOS

Ollama is the fastest way to run large language models on your own hardware. It manages model downloads, GPU acceleration, and a local HTTP server that speaks the OpenAI API format. On macOS, there are two install paths: a Homebrew formula for people who live in the terminal, and a native app for everyone else. This guide covers both.

What you need

• A Mac with Apple Silicon (M1/M2/M3/M4) or an Intel Mac with at least 8 GB RAM
• macOS 12 Monterey or newer
• Homebrew installed (optional but recommended)
• About 5 GB of free disk space for a small model

Option A: Install with Homebrew

Homebrew is the cleanest path because it handles updates with a single command. Open Terminal and run the two commands below. The first installs the Ollama formula from the core tap; the second starts the background service.

The service starts automatically on login. To stop it, run brew services stop ollama. To update later, run brew upgrade ollama.

Option B: Install the native app

If you prefer a menu-bar app over a background service, download the macOS installer directly from ollama.com. The .dmg contains a standard macOS application. Drag it to /Applications and open it. A small llama icon appears in your menu bar when the server is running.

Step 2: Pull and run your first model

The ollama run command pulls a model on the first invocation and then opens an interactive chat prompt. Llama 3.3 is Meta's flagship open model and a solid first choice. The 8B variant fits in 8 GB of RAM.

The model weights are cached in ~/.ollama/models after the first pull, so subsequent runs start in seconds. You can run ollama pull llama3.3 separately to download in the background without opening a chat.

Step 3: Verify the local server responds

Ollama starts an HTTP server on localhost:11434. You can check it is running and list downloaded models with a simple curl request. This is also the endpoint used by any client library or app that integrates with Ollama.

Managing models

Use the ollama CLI to manage what is stored on disk. Models live under ~/.ollama/models and can be several gigabytes each, so cleaning up unused ones matters.

• ollama list — show all downloaded models with size and modification date
• ollama pull mistral — download a model without running it
• ollama rm llama3.3 — delete a model from disk
• ollama show llama3.3 — print the model card (context length, architecture, license)

At this point Ollama is installed, a model is cached locally, and the API server is responding on port 11434. The next step is using the OpenAI-compatible endpoint from your own code, which is covered in the guide on using Ollama as a drop-in OpenAI API.