I ran out of disk space, so I shipped a paid app to curate a lifetime of photos.

The problem (it was mine)

I ran out of space on a 2TB laptop. I did a storage audit with Claude, which flagged a mountain of video and multimedia and suggested moving it to the cloud. My reaction: that is like moving a house I do not want into another house. The real problem was not storage. It was that a decade of photos and video was scattered across my phone, my desktop, and several drives, thick with duplicates and forgotten moments, and I had never actually curated any of it.

Apple Photos only manages what lives inside Apple Photos. Nothing I could find handled media across external drives, local folders, and iCloud at once, and then let me make fast keep-or-lose decisions without dread. So I built it.

What I built

KeepRoll is a local-first photo-curation app for the Mac. You point it at everywhere your media lives and it gives you one fast, calm place to curate it:

• Curate across every source. iCloud, external drives, and local folders in one library, not just what is trapped inside Photos.
• Swipe keep or trash. A fast, tender flow for making thousands of small decisions without it feeling like a chore.
• Surface the best of a burst. Beyond exact duplicates, perceptual-hash clustering groups near-identical shots so you pick the keeper and drop the rest.
• Share the work. Share links let someone else help curate a shared library (my wife curated ours).
• Sync keepers back. Your curated set syncs back to the iPhone, so the cleanup actually reaches the device you use.

Try it yourself

This is the actual KeepRoll interface, running live below on a sample library. On a computer, use the arrow keys or the on-screen buttons to keep, trash, or archive; on a phone, swipe. It is the same curation flow the real app uses.

A sandboxed, client-side demo. The real app runs on your own Mac across iCloud, external drives, and local folders.

How it is built

The engineering choice I am proudest of is restraint. KeepRoll is deliberately lightweight: a Python standard-library HTTP server with an embedded HTML/CSS/JS front end, no web framework and no build step, SQLite for state, and perceptual hashing rather than a machine-learning model for near-duplicate detection. The core is a 14-mixin Sorter architecture that keeps the logic composable.

I built it solo, with Claude Code, in roughly seventeen days across 239 commits. That velocity is the point: a real, shippable, paid product from a standing start in under three weeks, because the tooling let me move at the speed of the idea.

flowchart LR
  S["Scattered media
iCloud · drives · folders"]:::raw --> K["KeepRoll
dedup · perceptual-hash clusters · swipe"]:::mine
  K --> C["A curated library
keepers only"]:::good
  C --> P["Synced back
to your phone"]:::mid
  classDef raw fill:#fee2e2,stroke:#dc2626,color:#7f1d1d;
  classDef mine fill:#dcfce7,stroke:#16a34a,color:#14532d;
  classDef good fill:#dbeafe,stroke:#2563eb,color:#0b3a8f;
  classDef mid fill:#eef2ff,stroke:#6366f1,color:#1e1b4b;

Why it matters

This is the clearest example of how I work: take a real, felt problem, frame it sharply, and ship a product that resolves it, fast. Most people sit with a “what if I could…” and never build it. I turn that into a live, paid system in weeks. KeepRoll also shows range, it is a polished consumer product, not an internal tool: design, packaging, a marketing site, and a checkout, all the way to a paying customer. That instinct, problem to shipped product, is exactly what I bring to a team that has more good ideas than it has time to build.

By the numbers