Look at the file organizers that appeared in the last two years and you will notice most of them bill monthly, or bill once and then meter you: so many renames included, more if you need them.
That is not greed and it is not a growth tactic copied from someone else. It is arithmetic, and once you see it you can predict a tool's price from how it works.
An app built on a cloud model pays for every file you process, forever. It cannot sell you a one-time licence and then absorb your next fifty thousand documents, because your fiftieth thousand file costs it exactly as much as your first. A subscription is not a preference for that architecture. It is a requirement.
Some are open about it: you will see plans measured in renames per month, or an option to bring your own API key so that you pay the model provider directly instead of paying them. That is honest pricing of a real cost.
An app that runs the recognition on your machine has a different shape. The work is done by your own processor, using frameworks Apple ships with macOS. The developer's cost per file is zero. So it can be sold once.
Before you pay for anything, the question worth asking is not "how much" but "where does my document go to be read". The answer determines both the price you will pay over five years and where your bank statements end up.
Three ways to work it out:
Hazel has sold a perpetual licence for fifteen years — $42, one payment, free updates within the major version. It can do that because it never had a per-file cost either: rules are code running on your Mac. As of version 6 it can also run text recognition when a rule needs to match against the contents of a scan.
The trade there is not privacy or price, it is effort. You write the rules, one pattern per kind of document, and you maintain them when the documents change.
I am not going to pretend local is free of trade-offs, because it is not.
A model that runs on a laptop is small. It cannot reason about your document the way a large cloud model can. Anyone who tells you otherwise is selling something. The way to build on it honestly is to stop asking it to be clever: the text recognition reads what is physically on the page — reading is not guessing — and the model's only job is to choose which of the things already on the page should become the name. Extraction by code, selection by model. Get that inversion the wrong way round and you get invented dates and filenames with slashes in them, which is exactly what I got when I tried it the other way.
The second cost is your battery and your fans, on a big batch. The third is that it will never do the things that genuinely need a large model.
What you get in return is that the price stops. And that the documents never leave the machine, which for a folder full of statements, contracts and medical results is not a small thing.
Ask yourself what happens on the day you stop paying.
With a subscription tool, the organizing usually stops with it, and sometimes so does the search index that made your documents findable. What you were renting was the order itself.
The alternative worth looking for is a tool where uninstalling changes nothing: the files keep their new names because names are just names, the folders are ordinary folders, and everything stays exactly where it is. That is a property you can test before you buy, and it tells you more than any feature list.
The app I build is Tidy. $9 once, no subscription, no account, no server: the text recognition and the naming both run on your Mac, and you can verify that by turning off the internet. It never deletes anything, every run undoes in one click, and if you delete Tidy tomorrow your Mac stays exactly as tidy as it is today. Solo developer in Santiago, Chile.