Upload and discuss pdfs in composer

[BrianC (Brian)]

Hi everyone,

I’m not sure if this is the correct way to make a feature request, but I wanted to reference this post with Sam.

The idea is to allow users to upload PDFs or TXT files directly in chat and have the AI process and respond to questions about the content. It seems like much of the infrastructure is already in place, and it could add tremendous value for interactive discussions.

Would love to hear thoughts on this!

Thanks,
Brian

[MachineScholar]

This is a perfect solution for countless organizations who run a Discourse instance. Furthermore, the AI plugin could get a global setting to allow/disallow document uploads to personas and the admin panel could get a new tab with some settings for chunking size, number of results, etc. which can be global or per-LLM. As you mentioned, it seems like much or even most of the code infrastructure already exists for this feature.

@MachineScholar

I personally work with many people who can’t really use any AI in their work (ChatGPT, Claude, Perplexity, etc) because of privacy and other concerns. They also don’t have and can’t afford dedicated AI engineers to build a custom solution, nor are they technical individuals themselves. So I see a very large market gap here for small and medium businesses/organizations, where Discourse AI is the perfect solution.

[BrianC (Brian)]

@sam do you think this one will see development at some point?

[MachineScholar]

Hopefully there’s hope as we just got PDF uploads in the persona editor for RAG

PDF support in Discourse AI

This guide explains how to implement and use PDF processing capabilities within discourse-ai, including both basic text extraction and enhanced processing with LLM assistance. Required user level: Administrator Summary The discourse-ai plugin supports PDF processing for RAG (Retrieval-Augmented Generation) in two distinct modes: Basic text extraction Enhanced processing with LLM analysis Basic text extraction This mode provides fundamental PDF processing c…

[sam (Sam Saffron)]

The tricky thing here is UI.

There is a “lead time” fora PDF or large text file to get indexed. And there is a UI challenge.

1. Post #1,7,27 have uploaded PDFs/Text… how does user figure out what isgoing on?
2. You upload in post #1 but it will be 5 minutes till it is available, how do you communicate that to the user?

Technically, I think the only way around both that is straightforward is some sort of custom UI for uploads.

The UI challenge here is the biggest hurdle, the actual mechanics are not too complicated.

cc @awesomerobot / @Falco

[MachineScholar]

@sam

Technically, I think the only way around both that is straightforward is some sort of custom UI for uploads.

I’m not a frontend guy by any means so all of this may be nonsensical, but hold my beer…

AI upload modal

I propose that there could be an “AI doc” button in the composer’s toolbar which launches a special AI upload modal (via which the user can select standard or enhanced processing and some other settings). Then the topic/post can either be (and perhaps there could be a site setting to determine this behavior):

1. Greyed out and invisible to others in the thread except to the user posting until the processing is completed — after which, it becomes publicly visible. OR…
2. The post could be saved as a draft and the upload continues in the background. When opened it shows that AI upload modal with the upload progress. Essentially disallowing publishing until the AI upload is either finished or canceled.

• In both cases, there could be a setting which allows sending a notification to the user when their upload has finished.

Keeping track of uploads

In regard to tracking the uploads throughout the topic, a list of the files with their associated post number (plus perhaps size, etc.) could be injected in this location under the topic stats: