It was after noon on a Saturday when I found myself still in my PJs, sitting in a house that desperately needs a reset after a busy work week. Am I sick, or taking a well deserved lie in? Nope. I was glued to my desk all morning, setting up the Mac Mini that I had been stalking all week before it finally arrived on Friday, with the aim of turing it into my local AI development workhorse.

I am deliberately challenging myself to use a very different set of tools from the well funded Anthropic models I have access to at work, backed up by a suite of skills and harnesses carefully curated by our department. I weighed up a tradeoff between funding my personal AI usage in the cloud, and investing in a computer that can reasonably handle local models. I settled on the latter, choosing an M5 Pro with 64GB unified memory.

I have been working at the same company for 4 years, so I am not well practiced in setting up a development machine. Does AppleCare+ cover you if you throw it out of the window? Asking for a friend.

The stack I am using to run models locally is:

Ollama - a platform that downloads, manages and runs LLMS locally QWEN - a family of LLMS that can be run locally

Together these worked, but not well enough for coding - QWEN3.6 was being extremely verbose, refusing to read an .md file, and wildly hallucinating that I had asked it to write a function called get_fractal(). I clearly needed to set some parameters, and I needed something with pair programming capability. Enter Aider - an AI coding assistant.

Here is when I started to check my AppleCare policy - using the right terminal command, remembering all the required parameters, making sure Ollama is running, not running multiple instances of Ollama…… all sound like something that needs a script - so I wrote one, which I plan to build on once I start adding more parameters. You can find it here

Updated: