Is it possible to design a function similar to the project in Claude?

Because the knowledge base involved in ordinary chats or AI agents is too limited. If we want to conduct conversations using a knowledge base created from a large number of documents, it can be quite difficult without considering the issue of tokens.

In addition, in ordinary chats, if I upload documents, for example, I uploaded 10 conversations that have gone through two rounds, can I choose to delete some of the previously uploaded 10 for validity? This would improve flexibility and timeliness