The task
Paperless-ngx stores scanned documents, and each one needs tags, a document type, a correspondent, and a clear title before it can be found again. Before version 3.0, the AI help came from separate add-on containers. The author wanted to see what the AI built into the new release does with the author's own receipts and documents.
What they did
In version 3.0 the AI settings are a tab in the configuration. The author entered a language model, an embedding model, and an endpoint, and tested two setups: a cloud service through OpenRouter, and a local Ollama server in the home network. With AI switched on, every open document has a Suggest button in the side bar. One click and a short wait produce a list of proposals: tags, document type, correspondent, and title (the final release adds storage path and date). Each proposal is taken over separately with a click. The tests ran on a release candidate, and the author later updated the article for the final release. The same configuration also powers a chat over the whole archive that links to its source documents. This entry covers only the suggestions.
What the source reports
According to the author, one test document received 9 tag suggestions, including fitting ones such as tickets and event, and the document type and title were proposed as well. According to the author, the quality depends strongly on the chosen model: two cloud models did well, while very small local models missed correspondents and titles noticeably more often.
Where the approval sits
A person accepts every suggestion. According to the author, the AI changes nothing on its own: each tag, type, correspondent, and title is taken over by hand. New uploads are not classified automatically either, so a person has to open each document and click Suggest.
Worth knowing
This is one author's hands-on test with own documents, and part of it ran on a release candidate. With the cloud setup the text of the documents goes to an outside service. The local setup keeps it in the home network but is slower, and the author warns that it can time out on weak hardware. The article says version 3.0 has no switch to classify new uploads automatically, and it points to add-on tools for batch scanning workflows.
Source: mizine.de · Aug 2026
Drafted with AI assistance and checked automatically. A person is responsible for the collection. How these entries are made
Prompt for your agent
Suggest tags and sender for scanned documents
Suggest a title, tags, sender and document type for each scanned document in a batch, so you can check the list before you use it.
More in this category
- Back office and documents
A Paperless-ngx archive holds paperless-gpt suggestions for review
A hosting provider describes paperless-gpt, a service that proposes titles, tags, correspondents, document types, and dates for scanned documents in Paperless-ngx. Documents with the review tag wait in a screen where a person approves each change.
Read the entry - Back office and documents
A sole proprietor books receipts with Claude Code and approves each
A one-person business built a Claude Code skill that reads receipts from its Fiken inbox and proposes an account for each one from past bookings. The owner confirms every proposal before anything is booked.
Read the entry
Ninety minutes live on your tools. One automation running before we hang up.