[leobg]

Indexing Discord is going to be tough. The reason is that context is all over the place:

Question in one message. Then two unrelated messages. Then a partial answer by somebody. And so on.

It’s even worse than indexing a PDF. Just breaking stuff into paragraphs and generating embeddings isn’t going to cut it.

[sprremix]

I imagine this will only work (and only index) threads. So the context can be gathered from the thread title/body and underlying messages reflect the discussion.

Some communities I'm in have #support channels which only support threads. So you create a thread, add a title and a body message and people can reply to your thread by clicking on it. There's no way to post individual messages; only comments in threads.

Thread overview: https://i.imgur.com/jfvrRtG.png

Opening a thread: https://i.imgur.com/pqGrARI.png

This solves your context problem. Still not sure if this is the right direction we want to go in. This just proves to me that Discord is not right tool for the problem at hand.