Paperless-ngx

Paperless-ngx is the least glamorous and most durable tool in this directory. It watches a folder, OCRs whatever lands in it, files it, and makes it searchable. Ten years from now that is still what you want from a document archive.

The intelligence is deliberately modest: Tesseract for text, then a trained classifier that learns from documents you have already tagged. After a hundred invoices it starts suggesting the right correspondent and document type on its own. No language model, no GPU, about 2 GB of RAM and a single Postgres container next to the app.

Why it belongs on a low-spec list

Most "AI document" products are a chat box bolted onto a cloud OCR API. This one is the opposite: local, boring, and it degrades gracefully. On a 2-core machine the OCR step takes a few seconds per page and happens asynchronously, so you never wait for it.

If you add a language model later, the natural upgrade is a summarisation step on top of the archive — but the archive is the valuable part, and it works today without one.

Practical setup notes

• Set PAPERLESS_OCR_LANGUAGE to the languages you actually scan. Each extra pack slows OCR and increases misreads.
• The consume folder can be a network share, which is how most people wire a scanner or a phone scan app into it.
• Back up the media directory and the database dump. The database alone restores a list of documents pointing at files that no longer exist.

It does one thing, and it does it on hardware you already own. That is the whole review.