Right now the vector db is only initialized if the vector db scanner is used. because of this I can't easily add an API endpoint that indexes new embeddings.
The DB should be de-coupled from the scanner. DB gets initialized with the main apps, and the scanner is either passed a client to use or still pass the collection name and start a 2nd client for the scanner. One client is probably better for memory usage.
deadbits
commented
1 year ago
Right now the vector db is only initialized if the vector db scanner is used. because of this I can't easily add an API endpoint that indexes new embeddings.
The DB should be de-coupled from the scanner. DB gets initialized with the main apps, and the scanner is either passed a client to use or still pass the collection name and start a 2nd client for the scanner. One client is probably better for memory usage.