I understand that you can have an assistant with its own vector store with files and a threads with a similar vector store. Now if I interact with the assistant and thread A (each with their vector store) and then start a new thread B (with a new vector store but the assistant keeps its vector store), will the assistant have “memory” of any data discussed in Thread A?
In plain SRE terms - using the same assistant and treating Threads as tenants, will that work?
By my knowledge, Assistant while running on Thread B won’t have the memory of your conversation in Thread A.
Someone can correct me if am wrong.
More precisely - use of file attachments API methods creates its own vector store ID that is connected to a thread, expiring after 7 days. Those embedded documents then also are part of what can be returned by file search when interacting with the thread.
The store is unique to the thread.
It is persistent in the thread until expiration.
The return object from creating a message with attachment for file_search does not return the vector store ID, so you don’t have an easy path for management of this nameless ID (which can be listed out by GET https://api.openai.com/v1/vector_stores), apart from deleting the thread. You can be ultra-tricky and re-use this vector store ID, but being part of a user’s content, it could serve little purpose.