The discussion revolves around adding support for searching PDF files within forum topics via AI personas. Neil_Evans2 highlights the current limitation: PDF attachments aren’t indexed in semantic search, forcing users to manually upload them to RAG. They propose extracting and indexing PDF text during embedding generation to make content discoverable through AI search, eliminating duplication and simplifying the system. 2 Moin references a similar request in the meta forum, while pfaffman suggests a plugin could embed extracted text into posts, possibly as a hidden detail, and notes it could be developed as a PR. Neil_Evans2 adds that uploading PDFs to personas currently disrupts AI search for regular content, leading them to convert key PDFs to markdown or post summaries alongside files, and to switch to Sonnet 4.5 and disable HYDE for reliability.
Current limitation:
• Forum semantic search only indexes post text, not PDF attachments
• PDF files aren’t searchable via /search AI toggle
• To get round this, I had to manually upload PDFs separately to persona RAG
Proposed solution:
• Extract text from PDF attachments during embedding generation
• Index PDF contents alongside post text
• Make PDF-attached topics discoverable via semantic search
Benefits:
• Users find technical documentation via forum search
• No need to duplicate content (forum post + RAG upload)
• Better SEO (PDFs attached to indexed topics)
• Simpler architecture (Search command just works)
If you were to implement this, i could potentially:
Remove forced tools (Search would naturally find PDF contents)
Eliminate RAG uploads entirely (everything in forum topics)
I think that a plugin might add the text to the cooked post in an optionally hidden details element. That would add it so it would be found, I think. If you’re self hosted I think it’d be a only a few hundred dollars to have developed. Or, if it sounds like they’re interested, submitted as a PR, for about twice as much (to include tests and such).
Fyi - I found that uploading PDF files to the persona stopped it from finding “normal” forum content in the ai-assisted search. I have therefore resorted to a combination of (a) converting the key ones to markdown (so I can post them directly as topics) or (b) picking the main keywords/ToC etc. out and posting them alongside the PDF files in the forums. I also had to switch from GPT 4.1 to Sonnet 4.5 and disabled HYDE to make it reliable.