Music is terribly important to me, and so is owning my data. I don’t use streaming services, and over the last ten or so years, I developed a pretty bad habit of just dumping all new of my music in a “To Sort” folder. Flash forward to now, I have over 2000 albums and singles to sort through. This is particularly a problem due to legitimate limitations (or bugs) in network file access protocols like SMB and NFS, making accessing my music in the apps I use pretty unstable.

Over the years, I’ve had sprints where I’d get through as much as I can by hand but it usually results in many lost hours, general nausea and a really sore wrist.

Anyway, I was bored a few afternoons ago and revisited a project I’d started with Claude Code to help make sorting things easier. I don’t trust LLMs to sort it for me—I’m very picky—but a lot of the manual work of fetching metadata from discographies and aggregation sites doesn’t need to be done by hand.

I started this as a “what if” to see how close an LLM could one-shot a single-purpose app like this with very little in the way of guardrails or meticulous prompting. I got lazy with the first version and just ran it from my MacBook Pro, but quickly abandoned the whole project because there were simply so many files that macOS was choking loading them all over SMB. Once I deployed the app on my home server and removed all the Mac-specific hacks though, it ran just fine, so I took the very rough (and very bad) LLM designs and spent an afternoon cleaning things up and making a more pleasant experience.

I’ve spent the most time sorting in Random mode, where the app autoplays tracks from album as I’m sorting it. Sometimes I grab music completely arbitrarily or forget about it, so this has been leading to some pleasant surprises and reminds me of having access to Quick Look when I’d sort things manually in Finder.

While I still have a lot of very complicated feelings about LLMs that haven’t gone away, small projects like this do show how they might be useful. This is a tool for only me. I have no plans to open source it or develop it into anything bigger… but it does solve a problem I’ve personally had for a decade. I would never budget the time to make this, so how long (and how much wrist pain) would it take for me to sort through all 2000 items manually?

Would I have ever done it? Probably not.