Apple has refreshed the Mac mini and the Mac Studio. M6 is the company's first two-nanometer chip, and M5 Ultra is built from four dies and holds up to 512GB of unified memory at 1.2TB/s. Apple now talks about running large language models locally in plain terms.

What interests me is not the benchmarks. A box the size of a book, sitting quietly at home and running around the clock, is a ready-made server for an autonomous agent with remote access. Not for experiments, but for routine: triaging incoming tasks, assembling a sprint in Linear, preparing a summary for the morning.

The difference between the models matters. M6 tops out at 32GB, which is enough for smaller models and for an agent that calls external APIs. If you need a genuinely large model running locally, that is a Mac Studio with M5 Ultra.

A place for short notes on what I find worth keeping: products, technology, banking systems. Not articles, just observations, usually a paragraph or two.