Introduction
In this article, we’ll explore why I chose to use a M4 Mac Mini Pro as an AI workstation and how I have it configured in my setup . This approach allows me to leverage the power of MacOS while keeping things simple and efficient.
Why Choose a Mac Mini Pro?
Familiarity and Ease of Use
I’ve been using Macs for over 15 years, so I’m very comfortable with the operating system and its terminal interface. While Linux is also an option, my familiarity with MacOS makes it the quickest and easiest way to get things done.
Pre-Configured System
The Mac Mini Pro comes pre-configured out of the box, which saves time on initial setup. Additionally, it has a 12-year warranty, providing peace of mind in case of any hardware issues.
Compact Design
The Mac Mini Pro takes up very little desk space, making it an ideal choice for those with limited workspace. It’s also incredibly quiet, so doesn’t need hiding away or turning off if I’m recording audio.
Unified Memory Architecture
One of the standout features is its unified memory architecture. This allows the GPU to draw from the same pool of memory as the CPU, enabling larger AI experiments without the need for a separate, expensive NVIDIA GPU. The model of M4 Mac Mini Pro I went for is the 24GB RAM base model.
Setting Up the Mac Mini Pro
Static IP Address
Setting up a static IP address was one of the first steps I took. While DHCP is convenient, it doesn’t guarantee that an IP address will remain fixed indefinitely. By reserving a range of IP addresses on my home router for static allocation, I ensured that my Mac Mini always has the same IP address.
Screen Sharing
Setting up screen sharing between the devices means I can jump on the M4 Mac Mini to see what’s going on. This is easily configured in System Settings.
SSH / Remote Login
I also have SSH enabled which is located under Remote Login in System Settings. This allows me to remotely execute commands from a terminal running on my laptop.
Running Ollama
I installed Ollama on the local device and integrated it with a web interface called Khoj. This setup allows me to switch between different language models easily, read PDF files, access web content, and integrates with Obsidian for note-taking.
Benefits of This AI Workstation Approach
Flexibility
In conjunction with the static IP address I can access Khoj from any device on the home network through the browser, such as laptop and iPad should I want to. It can chew up the resources of the Mac Mini without impacting what I’m doing on my device.
Integration
The integration with Obsidian allows me to seamlessly access my notes and documents, enhancing productivity… I have the Obsidian Vault located in an iCloud folder, so again these files can be access elsewhere. On my iPhone I have 1Writer installed so I can edit the markdown files which are synced to iCloud.
Privacy
This setup allows me to run my own AI workloads in a completely private setup. I can test out a diverse array of language models and have a setup which best suits my needs.
Next Steps
My next endevour is to work out how to integrate N8N workflows and ComfyUI image generation. These are tools I use in the same way as Khoj (across the network), but it would be good to be able to send a prompt to Khoj to generate an image and this be taken care of Flux running through ComfyUI.
Conclusion
Using a Mac Mini Pro as an AI workstation has proven to be a practical and efficient choice. The combination of familiarity, ease of use, compact design, and unified memory architecture makes it an excellent tool for running AI experiments. By setting up a static IP address and integrating with useful tools like Khoj and Obsidian, I’ve created a versatile and powerful workstation that meets my needs perfectly.
Others may argue (perhaps correctly) an NVidia GPU would be a better solution, but for me along with learning the AI technologies I didn’t really want to be re-learning the art of building a PC too.
One More Thing…
This post was originally drafted as a voice memo / transcription and then converted into a blog post using Khoj and Mistral Small 3, to format and structure the post. There was some inevitable editing, but lots of time was saved…
The featured image was generated using ComfyUI having been fed the prompt
Artistic, abstract image of data flowing through an AI workstation.
The prompt was one of four suggestions from Khoj / Mistral Small 3.
Pretty neat, huh?
No Responses