eikek / docspell

Assist in organizing your piles of documents, resulting from scanners, e-mails and other sources with miminal effort.
https://docspell.org
GNU Affero General Public License v3.0
1.65k stars 127 forks source link

Feature Request: LLM-based Auto Tagging and Summerization #1996

Open pschichtel opened 1 year ago

pschichtel commented 1 year ago

I've just watched GPT-4's intro stream and this seems like a pretty good use case for it. ChatGPT could already be used for auto tagging based on the existing OCR results, with GPT-4 it might be possible to skip the OCR all together.

I'd be willing to submit patches for a prototype based on ChatGPT/gpt-3.5 some time in the coming weeks.

eikek commented 1 year ago

I imagine, this could be a good use case for an addon, that can hook into the processing of each file. The only thing that is highly discouraging for me personally is sending all my documents to chatgpt :) I think for the daily hassle papers, like invoices etc it is not so much useful, but if you use docspell for keeping track of papers and articles, this can be really nice!

LightTemplar commented 1 year ago

I could imagine using any sort of AI, but only if it hosted by me. So, using of ChatGPT is truly a case for an addon.

pschichtel commented 1 year ago

I've read up on the BLOOM model which is an open access model trained in the EU, which seems interesting. There is also some other (smaller) models that can feasibly run on a small home server. I have on my todo to get something like that running and to feed some OCR results into it.

pschichtel commented 6 months ago

I've just toyed around with the latest llamafile release and I have to say: I think we are at the point where it starts to be feasible to have a LLM running locally for auto tagging on import.