Content Embedding

We have to embedding content in new data first, and each content(or data) has one embedding vector.
When the user wants to start conversations, we will send all vector which relate to all chosen libraries(every library may have lots of content, i.e. lots of vector), into GPT.
The tool we use: Pinecone

No files, tests, or entry points are named in the issue. Clarify where content is stored and how conversations are started, then define the embedding flow using Pinecone; done should mean each content item has one embedding vector and selected libraries' related vectors are sent to GPT.

Written by the indexing model from the issue text.