I was looking forward to trying out OAI’s RAG implementation by attaching corporate onboarding documents and our Handbook to the Assistants I’m creating, but … I can’t. I discovered at the last minute that there’s a 20 file limit on Assistants. If the system is using embedding-based lookups, then there really shouldn’t be a limit like that.
Could the API dev’s please consider bumping up this number substantially? It doesn’t fit what is probably a very common use case, which is having corporate knowledge bases available to chat about without having to explicitly reference files, tools-style. The bot should just “know”, or have “memories”, of corporate knowledge.
(I already have this implemented, so for now I just have to keep my code active until when/if this is resolved)
You could get around this by meeting and using multiple assistants. E.g., an assistant to answer holiday and PTO related questions, another to help with timesheets and expensing, and so on.
You could then have an assistant to orchestrate these assistants. Once you give this orchestrator details about which sub-assistant does what, it can decide for itself what to use when.
Edit - just realised that my username is openai29. I’m not affiliated to OpenAI except as a third party developer.
The previous commenter simply comments all sorts of random stuff for whatever reason…
For the time being the only way to give it “more than 20 files” is to either do Embedding manually by code (look up LangChain) or condense you knowledge base into a few big files, and that “should” do it.
Thanks, yeah, I looked that up after responding.
I have my own RAG implemented, so I can stick with that. I will try creating a large txt file with the contents of those documents, see how it goes. Maybe one per category, as I have less than 20 categories.