Implement RAG

[svange]

Description

We have the ability to record conversations and to like/dislike parts of conversations. Each like/dislike (reaction) is stored in S3. These responses should be stored as embeddings in a vector DB. Langchain agents should be configured to use the DB for RAG.

Possible Solution

Implement a new AgentConfig option rag which hooks the agent up to a RAG vector store maintained by us.

Additional context

We must come up with a RAG strategy first.

1. One DB for everything?
2. One DB per customer?
3. One DB per OB user?
4. One DB per industry?

Additional context

No response