The new “assistants tools” with “knowledge retrieval” seems like a really cool complete RAG solution:
Assistants tools - OpenAI API
The only thing I don’t understand is how to deal with users who don’t have permissions to all uploaded documents.
Is it possible somehow to limit the search only to a subset of documents?
Thanks in advance,
I think the easiest way is to create different Assistants and give people access to the appropriate Assistant that has the right number of files attached.
I guess that could work if there aren’t many subsets.
But sometimes you want to use RAG on a content management system where each user may have a completely different subset…
If really on a ‘per user’ base you could add the files to a thread instead?
hi @jlvanhulst thanks for replying
I’ve done some tests in the playground and I’m not sure anymore this tool is good enough for searching a large set of documents…
first, only 20 documents can be attached, whereas a regular vector database can work on millions of records.
then the assistant doesn’t specify on which documents it found the answer, it just compiles an answer.
lastly, it also gets answers outside the knowledge base… :-\
I do not think Assistants were ever setup to be an alternative for huge retrieval RAG’s. They are also in beta at the moment.
The assistant will give you quotations - that you can map to the locations in the files, how many an how verbose is also depending on your instructions.
And your last point - very well possible - but also hard to say much about without your prompt.
hi @jlvanhulst, i now understand better.
really appreciate you help!! thank you