Permission requirements in paperless-ngx

[bjoernwuest]

Hi, I refer to this issue: #226

I did check the source code, basically the paperlessService and found the following usages against paperless-ngx-API:

Get information from paperless

GET /tags/
GET /correspondents/
GET /documents/
GET /user/

Updates

POST /tags/
=> e.g. to create the "ai-processed" tag on setup. Not sure if more tags would be created!

POST /correspondents/
=> used to create new correspondent derived by AI model

PATCH /documents/
=> removeUnusedTagsFromDocument
=> add tag to document (e.g. the ai-processed tag)
=> add correspondent, if there is no correspondent assigned to the document

This would result in the following privileges required:
Document: read and edit
Tag: read and add
Correspondent: read and add
User: read

Why I start this discussion?

• Paperless-ngx has its own processing to scan document and assign tags and correspondent (and document type) automatically. Why paperless-ai tries to assign correspondent too? (real question, would like to understang).
• To further limit privileges (and not sure if paperless-ai would assign additional tags beyond ai-processed), why not ask the user to specify an (existing) tag from paperless-ngx to use and disallow paperless-ai to create tags?

[bjoernwuest]

Just understood that paperless-ai also creates tags beyond "ai-processed". Yet, those tags are rather "ridiculous" (in my opinion). Some examples:

• IBAN
• Einzelpreis
• EUR
• Leistung
• Guten
• Grüßen
• Frankfurt

Some of those tags may be meaningful to some people (e.g. someone who wants to tag by location, e.g. Frankfurt), but I assume for most those tags are not relevant.

I removed the permission to create tags from the paperless-ai user and request that the paperless-ai does not create tags (and correspondents) unless someone explicitly requests to do so (i.e. via config flag).

[clusterzx]

Just understood that paperless-ai also creates tags beyond "ai-processed". Yet, those tags are rather "ridiculous" (in my opinion). Some examples:

• IBAN
• Einzelpreis
• EUR
• Leistung
• Guten
• Grüßen
• Frankfurt

Some of those tags may be meaningful to some people (e.g. someone who wants to tag by location, e.g. Frankfurt), but I assume for most those tags are not relevant.

I removed the permission to create tags from the paperless-ai user and request that the paperless-ai does not create tags (and correspondents) unless someone explicitly requests to do so (i.e. via config flag).

Then your model you are using is completly garbage or your prompt. Most users if not almost all users love the results and want exactly the ability to let it autogenerate matching tags.
I had NEVER such low result tags.

[bjoernwuest]

Ok, actually I did not forumlate any prompt against the AI but let paperless-ai just scan and index the documents. The model used is phi4 (less memory, better results than llama3.2)

[clusterzx]

Why I start this discussion?

• Paperless-ngx has its own processing to scan document and assign tags and correspondent (and document type) automatically. Why paperless-ai tries to assign correspondent too? (real question, would like to understang).
• To further limit privileges (and not sure if paperless-ai would assign additional tags beyond ai-processed), why not ask the user to specify an (existing) tag from paperless-ngx to use and disallow paperless-ai to create tags?

Most users I came accross here use the users that was created during paperless-ngx first setup or other super users. You can of course limit it to only the needed permissions.

To answer your questions:

Paperless-ngx has its own processing to scan document and assign tags and correspondent (and document type) automatically. Why paperless-ai tries to assign correspondent too? (real question, would like to understang).

• It is a core functions of paperless-ai to automatically create correspondents if not exist? Thats 50% of what it is about.

To further limit privileges (and not sure if paperless-ai would assign additional tags beyond ai-processed), why not ask the user to specify an (existing) tag from paperless-ngx to use and disallow paperless-ai to create tags?

• This is already possible? You can limit tags and also tell it to only use existing ones.

[bjoernwuest]

Ok, found the "ADD_AI_PROCESSED_TAG" and set it to no. Now, it seems to do not try to create tags anymore.