PDF processing

[humble sphinx]

Why is it that when I directly attach my .pdf to Google AI Studio, that it processes the PDF correctly while on the Libre side it gives me "Error processing file". Can I let my .pdf be processed by Google instead of whatever Libre is defaulting to?

[wide crystal]

LibreChat is defaulting to RAG, which is the system most AI Chatbots use (like ChatGPT). File processing by providers is planned and we will work on this soon

[humble sphinx]

Uh.. I think there is a slight confusion. RAG is not related to the processing itself. But I take it that there will be an update that fixes this soon :)

[rigid holly]

Hi @wide crystal , may i ask that the file processing by providers will incur extra cost to user that is charged by the provider? I imagine sth like provider has storage for the file and add processing layer to it (like RAG). Reason I m asking is that i setup local RAG using psql\pgvector with nomic-embed-text , and i begin testing with Claude desktop app file upload. Same model chosen, same question but the answer from Librechat and desktop app are completely different.

[wide crystal]

yes. I didn’t read the docs from Anthropic or Google but I’m pretty sure that the will definitely charge something

response are different based on questions. if you ask to summarize an entire PDF, imo full-context (like upload to providers or add the pdf’s text to the prompt is better, while for asking questions RAG works well too and costs way less

[rigid holly]

@wide crystal you are correct that in the summarization question, full-context approach has better response. Topic-focus (specific-matter) question deliver similar answers from both approach. Never tell a RAG-approach to summarize anything- paintful.