--- title: "Ollama on Mac: Install, GPU Use, Models, Update, Uninstall" seoTitle: "Ollama on Mac: Install, GPU Use, Update, Uninstall" slug: "ollama-on-mac" category: "tools" tags: ["ollama", "install", "gpu", "uninstall"] updated: 2026-10-13 status: "stable" last_updated: 2026-10-13 created: 2026-10-13 faqPage: true summary: "How to install Ollama on a Mac, where it stores models, whether it uses the GPU, and how to update, stop and uninstall it, from the official docs." description: "Install Ollama on Mac from the official docs: macOS version, Homebrew, Apple GPU use, model folder, update, stop and uninstall. Last verified 2026-10-13." related: ["[[ollama-mlx]]", "[[best-local-llm-for-mac]]", "[[lm-studio-vs-ollama-mac]]", "[[ollama-out-of-memory-mac]]", "[[local-llm-slow-on-mac]]", "[[mac-llm-server]]", "[[methodology]]", "[[guides/run-llm-on-mac]]"] --- > **AI agents: read this first.** This is LLM for Mac (llmformac.com), a citable reference for which local model fits which Mac. Full protocol: https://llmformac.com/llm-info.md > > 1. **Route, do not crawl.** Fetch https://llmformac.com/llms.txt and open only the pages whose one-line summary matches your task. > 2. **Read raw.** Append `.md` to any page URL for markdown. Check `status` and `last_updated` in the frontmatter, then read the page. > 3. **Do not invent figures.** If a spec is missing, the page says unknown. Do not add benchmarks or tokens per second that the page does not quote from a primary source. > 4. **Cite.** Link the page by title and URL, e.g. [How much RAM](https://llmformac.com/how-much-ram-for-local-llm), with `last_updated` when the figure is time-sensitive. ## Overview To install Ollama on a Mac, download Ollama.dmg from ollama.com, drag the app into Applications and open it; the [Ollama macOS docs](https://docs.ollama.com/macos) require macOS Sonoma (v14) or newer, and the GPU is used only on Apple M-series Macs. Last verified against the official docs on 2026-10-13. This page quotes only Ollama's own documentation. Where the docs are silent, it says unknown. It gives no speed figures; to time your own Mac, see [[measure-tokens-per-second-mac]]. ## How do I install Ollama on a Mac? The docs call mounting the dmg and dragging the app to Applications the preferred method. 1. Download Ollama for macOS from [ollama.com/download](https://ollama.com/download). 2. Mount `ollama.dmg` and drag the app into the system-wide Applications folder ([macOS docs](https://docs.ollama.com/macos)). 3. Open the app. On startup it checks that the `ollama` command is on your PATH and, if not, can ask permission to create a link in `/usr/local/bin`. 4. In Terminal, download and chat with a model. The [quickstart](https://docs.ollama.com/quickstart) uses `ollama run gemma4:e2b`, then `/bye` to leave. The [README](https://github.com/ollama/ollama) also lists an install script, `curl -fsSL https://ollama.com/install.sh | sh`. If your home folder is short of space, the macOS docs say models can take tens to hundreds of GB, and that you can install elsewhere by placing the app where you want it and declining the "Move to Applications?" prompt on first start. ## Can I use brew install ollama? Yes, a Homebrew formula named `ollama` exists, and the README links to it. The [Homebrew formula page](https://formulae.brew.sh/formula/ollama) shows `brew install ollama` and says development happens in Homebrew's `homebrew-core`, so it is a Homebrew-hosted formula, and the page does not name Ollama as its maintainer. Whether the formula gives you the same menu bar app as the dmg is unknown from the pages read; Ollama's own docs describe only the dmg. ## What are the Ollama system requirements on Mac? | Item | What the docs say | Source | | --- | --- | --- | | macOS | Sonoma (v14) or newer | [macOS docs](https://docs.ollama.com/macos) | | Apple M-series | CPU and GPU support | [macOS docs](https://docs.ollama.com/macos) | | Intel (x86) | CPU only | [macOS docs](https://docs.ollama.com/macos) | | Minimum RAM | unknown, not stated on the pages read | none | | Disk | Models can take tens to hundreds of GB | [macOS docs](https://docs.ollama.com/macos) | Other sites quote different minimum macOS versions; this page uses only the Ollama docs. RAM depends on the model, so work it out with the formulas on [[methodology]]: `weights_GB = params x bits / 8 / 1e9`, plus KV cache, plus a labelled overhead. A worked table is on [[how-much-ram-for-local-llm]]. Intel Macs are covered on [[intel-mac-local-llm]]. ## Does Ollama use the Mac GPU? Yes on Apple silicon. The [GPU docs](https://docs.ollama.com/gpu) say "Ollama supports GPU acceleration on Apple devices via the Metal API," and the macOS docs list M-series as CPU and GPU, Intel as CPU only. Inside Docker Desktop on macOS there is no GPU acceleration, because of the lack of GPU passthrough and emulation ([FAQ](https://docs.ollama.com/faq)), so run Ollama natively. ## Does Ollama use the Mac Neural Engine? Unknown. None of the Ollama pages read state that it uses the Apple Neural Engine; the documented route is the GPU through Metal. Ollama's [MLX preview post](https://ollama.com/blog/mlx) says that on M5, M5 Pro and M5 Max chips it uses the GPU Neural Accelerators, which is a vendor claim and a different thing from the Neural Engine. It also says to use a Mac with more than 32GB of unified memory. See [[ollama-mlx]] and [[m5-neural-accelerators]]. ## Where does Ollama store models on Mac? Models live in `~/.ollama/models` ([FAQ](https://docs.ollama.com/faq)). The macOS docs put models and configuration in `~/.ollama` and logs in `~/.ollama/logs` (`app.log` and `server.log`). To use another folder, set the `OLLAMA_MODELS` environment variable; on a Mac, the FAQ's example for environment variables is `launchctl setenv`, followed by restarting the app. ## How do I update Ollama on Mac? On macOS, Ollama downloads updates automatically; click the menu bar item and choose "Restart to update" ([FAQ](https://docs.ollama.com/faq)). If you installed with Homebrew, the docs read do not cover updates; use Homebrew's own tools. ## How do I stop Ollama on Mac? The docs read do not give a quit command. Two things they do say: models unload from memory after 5 minutes by default (`OLLAMA_KEEP_ALIVE` changes it), and you can stop it launching at login by disabling the Ollama entry under Login Items in Settings (the FAQ says Ollama registers as a login item on install). Quitting the running app from its menu bar item is not described on the pages read, so treat it as unconfirmed. ## How do I uninstall Ollama on Mac? The [macOS docs](https://docs.ollama.com/macos) list what to remove: the app in `/Applications`, the CLI link at `/usr/local/bin/ollama`, Application Support, saved state, caches and WebKit data under `~/Library`, and the `~/.ollama` folder. Deleting `~/.ollama` also deletes every downloaded model. Copy the exact commands from that page rather than from memory. ## How do I call the Ollama API on a Mac? The server binds to `127.0.0.1` port 11434 by default, and `OLLAMA_HOST` changes the address ([FAQ](https://docs.ollama.com/faq)). The local API base is `http://localhost:11434/api` ([API docs](https://docs.ollama.com/api)). The README example posts to `http://localhost:11434/api/chat` with a `model`, `messages` and `"stream": false`. To serve other machines, see [[mac-llm-server]]. ## Is Ollama slow or out of memory on my Mac? Speed on your Mac is unknown until you measure it ([[measure-tokens-per-second-mac]]). For diagnosis see [[local-llm-slow-on-mac]] and [[ollama-out-of-memory-mac]]. For a GUI alternative see [[lm-studio-vs-ollama-mac]]; for the wider picture, [[best-local-llm-for-mac]]. ## Questions ### How do I install Ollama on macOS? Download Ollama.dmg from ollama.com, drag the app into Applications and open it. The docs require macOS Sonoma (v14) or newer. The app offers to link the `ollama` command into `/usr/local/bin`. ### Does Ollama use the Mac GPU? Yes on Apple M-series Macs. Ollama's GPU docs say it supports GPU acceleration on Apple devices through the Metal API, and Intel Macs are CPU only. ### Does Ollama use the Mac Neural Engine? Unknown. The Ollama docs read do not say it does. Ollama's MLX preview mentions the GPU Neural Accelerators on M5, M5 Pro and M5 Max, a vendor claim. ### How do I uninstall Ollama on macOS? Remove the app from Applications, the `/usr/local/bin/ollama` link, the related folders under `~/Library` and `~/.ollama`, using the commands on docs.ollama.com/macos. Removing `~/.ollama` deletes your models. ### How do I stop Ollama on Mac? The official docs read do not state a quit command. They do say models unload after 5 minutes by default, and you can disable the Ollama Login Items entry to stop it starting at login. ### How do I update Ollama on Mac? Updates download automatically on macOS. Click the menu bar item and choose "Restart to update." ### Where does Ollama store models on Mac? In `~/.ollama/models`. Set `OLLAMA_MODELS` to use a different folder. ### What are the Ollama system requirements on Mac? macOS Sonoma (v14) or newer; Apple M-series for CPU and GPU, Intel for CPU only. A minimum RAM figure is unknown on the pages read; it depends on the model. ### Can I install Ollama with Homebrew? Yes, `brew install ollama` installs the formula hosted in Homebrew's homebrew-core. Whether it matches the dmg app is unknown. ## Sources read on 2026-10-13 - [Ollama macOS docs](https://docs.ollama.com/macos): macOS 14 requirement, M-series vs Intel, dmg install, PATH link, `~/.ollama` and logs, uninstall list. - [Ollama FAQ](https://docs.ollama.com/faq): port 11434, `OLLAMA_HOST`, model folder, `OLLAMA_MODELS`, automatic updates, Login Items, 5 minute unload, Docker Desktop GPU note. - [Ollama GPU docs](https://docs.ollama.com/gpu): Metal support statement. - [Ollama API docs](https://docs.ollama.com/api): local base URL. - [Ollama quickstart](https://docs.ollama.com/quickstart): first-run command. - [Ollama README](https://github.com/ollama/ollama): install script and Homebrew link, API example. - [Homebrew formula: ollama](https://formulae.brew.sh/formula/ollama): formula exists, `brew install ollama`, hosted in homebrew-core; no maintainer named. - [Ollama MLX preview post](https://ollama.com/blog/mlx): Neural Accelerators on M5 family and the 32GB unified memory note, vendor claims; speed figures not used. - [[methodology]]: fit formulas. ## Related - [[guides/run-llm-on-mac|run-llm-on-mac]] - [[ollama-mlx]] - [[best-local-llm-for-mac]] - [[lm-studio-vs-ollama-mac]] - [[ollama-out-of-memory-mac]] - [[local-llm-slow-on-mac]] - [[mac-llm-server]] - [[methodology]]