How to Update Custom Knowledge Base Files When Using a Vector Database with File Search

Prompting
, , , ,
1

Hi everyone,

I’m currently working on a project that involves using a vector database in conjunction with a file search mechanism to manage and update a custom knowledge base. I’m a bit unclear on the best practices for updating the files within the knowledge base. Specifically, I would like to know:

  1. How should I approach updating individual files or records within the vector database to ensure consistency?
  2. If I have a file that has been updated, how should I upload the new file, especially when the update only includes a small section stating that “XXXX has been updated into …”?

Sometimes, after uploading such an updated file, the model continues to provide the old information instead of recognizing the update. I’m wondering:

  1. Does this issue relate to the file name or the content within the file?
  2. Are there specific keywords or instructions that need to be included in the assistant’s prompt to ensure it understands that it should reference the new information from the updated file?
  3. Or there are something else that I should notice?

Any advice or pointers to documentation would be greatly appreciated!

Thanks!

2

Hi there and welcome to the Forum!

Just to understand better: are you working with the Assistants API and the vector store or are you using an external vector database?

4

Hi! I am using the Assistants API and the vector store.