CASE STUDY / BUILT WITH BASAL
Local document classification in Paperless-ngx
English translation of a contributor’s story. Read the Polish original →
I built an AI backend integrating Basal 1.5B with a home document archive in Paperless-ngx. From OCR text, Basal proposes tags, a correspondent, a document type and a storage location, choosing from existing values supplied by Paperless. A separate Bielik model generates the title. The user approves suggestions in the standard Paperless interface, without modifying its code.
Everything runs locally on a Mac mini M2 Pro, without sending documents to cloud services. In tests on invoices, a receipt and a contract, Basal classification took around 1.5–2.5 seconds after warm-up.
Submitted by a model user and published after moderation.