Using a vector db for Assistant API

any thoughts if it’s possible that instead of manually uploading a file in Assistants API for the retrieval tool, it will use a vector db to retrieve answer from the vector db?

You can surely implement your own vector search on your server and use a function for the assistant to have access.

But if you go that far I question whether it is worth persevering with the assistant API (and its relatively high cost) and just use the normal completions API (that also supports functions).