[RAG] What is the best way to add RAG in Agent?

[douyh]

Initial Checks

• [X] I have searched Google & GitHub for similar requests and couldn't find anything
• [X] I have read and followed the docs & demos and still think this feature is missing

Description

I can see some examples of rag in examples.

I have a question. I don't want to call rag every time, that is, I don't want to add ref_doc directly beacuase of huge time comsumption. How can I decide what is the right time to use rag? Use rag as a tool?

I hope to get some suggestions. Thanks.

Use case

No response

[zzhangpurdue]

The best practice is to index file once, and save indexing file, there is no way to avoid the indexing step which cost huge time, we default set the 'use_knowledge_cache' to True, to allow use cached indexing. BTW, we are upgrading our RAG ability, by using llama-index, user could easily apply the llama-index component in our agents.

[suluyana]

Using rag as a tool may be a good idea, but the effect may be questionable. Or call llm first to decide whether the current round of query uses rag.

[douyh]

The best practice is to index file once, and save indexing file, there is no way to avoid the indexing step which cost huge time, we default set the 'use_knowledge_cache' to True, to allow use cached indexing. BTW, we are upgrading our RAG ability, by using llama-index, user could easily apply the llama-index component in our agents.

I see. I will try llama-index. Thanks for your reply.