Vector database should be a persisted volume when using Docker so you don't have to download and load the data every time. Also, any auto_update added records will be lost between runs right now.
Modify entrypoint process so other embedding models can be used. right now its just openAI because they are hard-coded in entrypoint.sh