millfolio blog

Shaping your dashboard with a blind AI — typed widget results, a versioned board spec, and pages, each layer validated before it renders.

A from-scratch GPU inference engine written entirely in Mojo — what crossing Mojo 1.0 and its batching Metal driver bought in performance, why one language end to end, and honest numbers against MLX.

How millfolio keeps thousands of records queryable with an on-device model — a three-flavor tag system, batched classification with priorities, and previews that show you what a tag would actually match.

How millfolio runs a frontier model over your personal files without the model ever seeing them — and why the backend is written in Mojo.