Paperless-ngx Tools

This repository contains a set of scripts I've developed to streamline my document management workflow in paperless-ngx. I published them here just in case anyone finds them useful.

• src/create-titles.ts:
  When dealing with scanned PDF files, I often encounter generic filenames like "document.pdf." This script addresses that issue by generating more descriptive titles. It leverages ollama to analyze the document content retrieved from paperless-ngx using a local LLM (Large Language Model), then updates the document in paperless-ngx with the new title. This works surprisingly well.

• src/split-pdfs.ts:
  My scanner saves all documents as a single PDF due to its Auto-Document-Feeder feature. To tackle this, I utilize Stirling-PDF to split these PDFs into individual pages, which are then saved in a folder. This folder can then be conveniently uploaded to paperless-ngx.

Setup

Setting up these tools is straightforward. Simply clone this repository and execute yarn to install the necessary dependencies.

Create Titles

To utilize the title creation script, ensure your ollama API server is running, preferably with GPU access. For a quick setup, you can use docker. Additionally, gather your paperless-ngx URL and credentials, as the script interacts with paperless-ngx. Store these details in the config.json file. Then, execute the script using yarn create-titles.

An essential aspect affecting title quality is the LLM Model and Prompt. You can customize the prompt in the create-titles.ts file and adjust the model within the config.json file.

If you want to generate titles in languages other than German, make sure to adjust the prompt accordingly.

Edit this prompt to match the language of your documents, you can also adjust the example to give the model a rough idea of the expected title.

Since my documents are in German, I've opted for the "mistral" model as it delivers satisfactory results quickly. For English documents, a smaller model may suffice, while for other languages, experimentation may be necessary with different models.

Split PDFs

For the PDF splitting functionality, ensure Stirling-PDF is running and configure the URL in config.json. While there may be simpler methods available for splitting PDFs, I personally prefer the functionality provided by Stirling-PDF. create two folders, scans and documents, and place the PDFs to be split in the scans folder. Then, execute the script using yarn split-pdfs.