I’m using the Gemini API (2.5-pro) to parse through a large set of PDF files hosted on URL links. I’m not hosting them locally or on a VDB. Does Gemini have a sort of cache for URL lookups? Like imagine I have a file; it gets converted into a vector store, which I can use cosine similarity on for lookup (RAG-based model). Now, every time I execute my request, will Gemini have to convert it into a vector store again, or will it use the old one? My question is if Gemini has to keep doing this every time it references the same URL because latency is beginning to become a problem in my multi-agent system.
If I do need to use a VDB, my enterprise uses Weaviate, but the documentation hasn’t been updated for Gemini since the 1.5 models, and I lose the non-text information since I have to use strictly text with the tex2-vec-openai transformer (I don’t know if there’s a good transformer outside of it).
(Sorry if any of the terms I used were wrong, but you get the gist. I’m aware it’s an experimental feature. Also, why do they use discord’s emoji system here???)
