The community members are discussing the relative strengths of the chat_engine and query_engine in a system. Some community members suggest that the chat engine may be weaker, while others note that it depends on the specific chat engine used and that there are tradeoffs between speed and accuracy. The discussion covers various approaches to incorporating chat history into a query engine, such as using a retriever directly with a context chat engine, or wrapping the retriever in a tool and using it with an agent. The community members also mention that the documentation provides details on the available chat modes that can be initialized from the constructor.
Yah, cause most of the docs mention RetriverQueryEngine and I can’t find anything chat equivalent. I just want to have previous context sent in to create a basic chat experience.
"I just want to have previous context sent in to create a basic chat experience" -- turns out when doing that on top of RAG, there's many approaches to do that, and none are perfect 😄
You can use that retriever directly with context chat engine, or condense plus content context chat engine.
You could also wrap the retriever in a tool and use it with an agent. Or put the retriever into a query engine and also use with an agent
