Can assistant and thread share the same vector store in the same run?

API
,
1

Hi there, I’m creating an assistant and thread which needs to compound knowledge from uploaded documents over time.

In a single thread, files will be periodically uploaded attached to a message and processed with the assistant’s instructions.
As more files get uploaded, the assistant needs to continuously reference the latest document and the previously uploaded documents and history for that thread for its responses.

I also need to create new threads to be able to discuss the uploaded documents separately, hence why I want the assistant to also link to the vector store of the original thread.

Will this cause any issues sharing the vector store in the same run? Or will it trigger a double query of the same vector stores? Not sure what the implications are…

Your assistance or recommendations for the best way to approach this is very much appreciated. Thank you.

2

The result is duplicate returns.

Imagine a vector store with just file containing “[This is a chunk from a vector store]”

Creating an assistant with a vector store, and then attaching a thread vector store.

image
image694×451 8.76 KB

3

Hi J,

Thank you for verifying with that test. I’m still learning so your help is greatly appreciated.

If i’m not going to be creating citations/referencing the documents like you have done, is there any downsides for my use case?

4

Basically: you would get duplicate top-10 results instead of top-20 results.

You pay twice as much for the quantity of information provided, and the AI may be overfitted to where it likes to just repeat back from the results when it does a search.

1 Like
5

Thank you.

When creating new threads which need access to the same documents, instead of using the assistant as the link to the original thread vector store, seems like my best bet is to just link the new threads directly to the original thread’s vector store, leaving the assistant unlinked.