Introduction
DeepSeek is all the rave right now. Here’s a quick guide on how to get it running locally on your Mac. Here’s a YouTube walkthrough as well:
Download Ollama
Ollama is a desktop utility that lets you run several open source LLM models, including the Llama models by Meta. It added DeepSeek models recently.
To install Ollama, go to https://ollama.com/download and download the binary for your OS. I will focus on Mac. Double click the downloaded .zip file and drag the Ollama app icon into your /Applications folder (via FInder). Then open the app and these sequences should open up. It will also ask you to run > ollama run llama3.2 — which is Meta’s model and good for comparisons.
Run Ollama in Command Line
After a bunch of scripts and downloads, Ollama should be installed and automatically launches Llama v3.2. Note that it runs in the “command line” out of the box. Don’t worry, we’ll get your a “WebUI” later on. For now, you can start chatting with Llama for a test run, e.g.:
The math seems right. So the >>> is simply your “prompt input box.” You can type /? for non-prompt commands you can run. For example:
>>> /show info — this will show you the model you are running
>>> /bye — this will exit the model and Ollama
When you want to run Ollama again, you can open up Terminal app and then just type in > ollama — the following should pop up. You need to add an option argument.
Some handy ones would be:
> ollama list — this lists the models you have installed. For now, you only have Llama.
> ollama run <model_name> — I would run list above first and then type in the exact model name to start it (otherwise Ollama might go online and grab another simliar named models)
Install DeepSeek
Ollama supports a few DeepSeek distillation models (see https://ollama.com/library/deepseek-r1). For now, we can try the 8b one which is based off of Llama and is small enough to run on most Apple Silicon machines (M1 to M4). You can run the following command to install the other models later.
To install, in Terminal, type > ollama run deepseek-r1:8b — this will download the model and start running it. You can start asking it questions. I asked “why can’t chickens fly?” below.
It shows all the reasoning steps DeepSeek is asking itself (inside the <think> </think> tags), before giving the final answer at the end. That’s a main reason why many people are excited, as OpenAI doesn’t quite show you what’s under the hood too much.
Install Open WebUI (pre-requisites)
If you are still here and not lost by the command line (CLI), but prefer to run things in the web browser, here’s what you can do next. These steps also require CLI. These instructions are also on the Open WebUI GitHub page. On there, there’s an alternative method - via Docker.
pyenv + python
First, you need to get python and pip. For Python, one option is to do it via pyenv since it manages different python versions.
> brew install pyenv
> pyenv install 3.12.8 (there are many versions… you can check here https://www.python.org/doc/versions/ or run > pyenv install -l for a full list)
This should get you going. If you end up installing more than one Python versions, then you can do > pyenv versions to see them all and which is the default; and > pyenv global <version> to assign one to be the global default.
pip
pip is a python package manager (there are many other options, but this is the OG). It should be installed already for you after the above python installations. You can type > pip to check.
Install Open WebUI (for real now)
Open WebUI is a comprehensive project that allows services to run in web interface / browser.
To install — > pip install open-webui
To run — > open-webui serve
(Note: Do make sure that Ollama is running, either in another Terminal window, or you can click the Ollama Mac app. Otherwise, you won’t see any models in Open WebUI.)
A bunch of stuff starts running and you should see the following at the end:
Run Open WebUI
Now, go to your web browser and type in http://localhost:8080 (or just click this). It will first ask you to create an admin account — just fill things in.
And finally, you should see this screen and can talk to any installed models just like on ChatGPT website. The top dropdown lets you switch models.
After you sends a prompt and click the <thinking> dropdown, you can see the reasoning DeepSeek goes through as well.
When you are done, go back to Terminal and type Ctrl-C — this should terminate Open WebUI. You can quit the Ollama app as well. To run it again next time, you will simply 1) open the Ollama desktop app, 2) run > open-webui serve and 3) open http://localhost:8080 again.
Reference
The DeepSeek paper is an interesting read: https://arxiv.org/pdf/2501.12948