PROSUMER.ONE / SHOWROOM / NO CHECKOUT
Not a shop. Not a shortlist. A window of personal AI machines that teach us something about the category — including machines we would not buy.
A 3.5-litre AI lab whose software surface became more interesting than the headline GPU.
A 300 g-class bet on always-on private inference, with the converter pipeline as the real product dependency.
These live on PROSUMER.WATCH as candidate or research dossiers.
192 GB of capacity and a headline that needs deflating before it needs benchmarking.
A repairability philosophy with a soldered heart and a real-but-workful clustering path.
Two memory pools, one placement engine, and a question more interesting than GB per pound.
A shared accelerator over the network where API, SDK and virtual-driver promises must stay separate.
A finished-personal-AI bet where persistent memory and bounded actions matter more than peak benchmark numbers.
A machine can be important to the map without being right for me.
That distinction is the point of separating the showroom from the live index. PROSUMER.ONE explains the world in human terms; PROSUMER.WATCH keeps the evidence moving.