Pull statements from Paperless-ngx (or another document archive) by tag

[halvar20000]

Proposed by the first person to write in, who has years of statements digitised in Paperless-ngx — scanned paper, downloaded PDFs, spreadsheets — tagged by bank and by type. The ask: let the dashboard pull them itself, so hundreds of documents arrive without a single manual upload.

What Paperless offers

I have read its API, and the pull side is straightforward:

• token authentication (Authorization: Token …), the token made under My Profile;
• documents listed and filtered by tag, correspondent or full-text query (/api/documents/?…);
• the original file (/api/documents/{id}/download/);
• the OCR'd text of every document in the document JSON (content) — which matters for scanned paper, see below.

The shape I have in mind

1. Settings → Document archive: the Paperless URL and a token. Stays on your LAN; nothing else changes.
2. On each account: which documents are its — a tag (e.g. bank + statement), a correspondent, or a query. Several accounts, several filters.
3. Pull on the daily sync, and a Pull now button. Each document goes through exactly the recognisers an upload goes through (the same code as dropping a file on the import page). Every Paperless document id is remembered, so a re-pull costs nothing and nothing is imported twice.
4. What could not be read is listed by document, with a link back to it in Paperless, rather than silently skipped.

The honest limit

The pull is generic; the reading is not. A document becomes transactions only when there is a reader for its layout. Today that is: DKB (Girokonto/Tagesgeld/Visa CSV, Wertpapierabrechnung PDFs), Swissquote and Yuh statement PDFs, Crédit Agricole (Suisse), Trade Republic, Degiro, Finary CSVs, payslips, and any other CSV through the column mapping. A PDF from a bank the app has never seen is not read — pulling it changes nothing about that.

So the useful input for this idea is which banks and document types are in your archive. A redacted sample — or just the text layer of one PDF — is what a reader is written from.

The interesting part: scanned paper

A PDF that is only an image has no text for the readers. Paperless has already OCR'd it. Feeding Paperless's content to the same readers might just work for a statement whose layout is known — OCR keeps line order well enough on a clean scan. That is worth trying with one real example before promising it.

Not in scope

• Writing to Paperless (tags, notes) — read-only, like everything else the app connects to.
• Replacing the import page. Uploads keep working exactly as they do.

👍 this thread — or the comment linking it from the roadmap — if you would use it, and say which archive and which banks.