Every scan gets a name.
And a folder.

GoodSort gives scanned PDF files consistent names and files them into an existing folder structure. A local language model reads the text of every document and proposes a date, a counterparty and a title. You review each proposal on its own and either confirm it or change it.

Every correction is remembered and applied at once to all pending documents from the same counterparty. Five misread Telekom invoices therefore take a single correction.

Text recognition and language model run entirely offline. A full run works in airplane mode, which is the dependable proof. No document and no excerpt leaves the machine.

A run reads, recognises text and proposes a name and a destination. It moves no file at all. Only your confirmation in the review pane moves or copies anything.

Moving clears the source folder. Copying leaves your old filing completely intact and needs twice the space for it. You choose the mode before the run.

Each session writes its own file, one line per move, straight to disk. From it a single move or an entire session can be undone. Moved files return to their old place, copies go to the trash.

GoodSort finds byte-identical and text-identical duplicates and clears them into a folder of their own. Nothing is deleted.

A confirmed assignment applies to every further document from the same counterparty. If you change a name during review, a name equation is created, say UniCredit Bank AG to Hypovereinsbank. All pending documents bearing the old name receive the new one at once.

GoodSort reads your existing filing once. Only those destinations are then offered. The program invents no new folders.

The file name is built from blocks, by default {datum} {gegenueber} - {titel}. Empty blocks vanish together with their separators. The same holds for the destination path.

German and English ship with the app. Every visible word lives in a single JSON file. If you need another language, you edit the right-hand column there, without rebuilding the program.

1. Set up. Choose the source folder with your scans, choose the destination folder, read in the existing structure.

2. Start a run. GoodSort reads, recognises text, finds duplicates and proposes a name and a destination. The run can be paused and resumed at any time.

3. Review. One document after another. The proposal on the left, the page preview on the right. Return accepts and moves on, U undoes the last move.

4. Look back. Every move is written to the log beside your filing and can be read there or undone.

Long runs can also be started without a window, overnight for instance.

macOS 14 (Sonoma) or later. Text recognition and PDF tools sit inside the program, there is nothing to install for them.

The only additional software is ollama, a local model server. It is free and available at ollama.com. If it is missing, GoodSort points to it at launch and fetches the language model at the press of a button.

Missing something, hit a snag, or just enjoying GoodSort? I am glad about every reply, a few lines are plenty.

Send feedback

Concept and development: Jan Gutzeit. The app is offered free of charge. GoodSort is the sister program to GoodEdit and follows the same design.

One candid point belongs here: besides the queue, the working database also holds the recognised raw text of the first pages. It therefore carries the same confidentiality as your filing itself. Once the work is done, that text can be cleared out again with a single command.