i'm working on a RAG app using

At a glance

i'm working on a RAG app using LlamaIndex / llamacpp server / vectorpg / flask / react+ typescript chat UI

i've got a prototype working that:

uses a custom VectorDBRetriever to retrieve first batch of nodes with scores
response_syntheziser with custom prompt text_qa_template
RetrieverQueryEngine that uses SentenceTransformerRerank
and finally generating response via CondenseQuestionChatEngine.

problem: i'm unable to figure out how to roll in chat history so that the app functions like a true chat bot.

are there any documents / tutorials / code bases that would be a good reference to help me figure out this last piece?

4 comments

LLogan M

The condense question chat engine is already using chat history (abeit maybe not the best way)

We have several chat engines/chat modes though

LLogan M

IMO an agent will be the most customizable
But heres all the modes
https://docs.llamaindex.ai/en/stable/module_guides/deploying/chat_engines/usage_pattern.html#available-chat-modes

JJames

will revisit the chat modules. i remember having issues trying to use the simple chat one with my reranker.

and i've been meaning to look into agents, will dig in!

JJames

and maybe condense question isn't the best but it was the only one i could get working. had trouble with context chat as well but i don't recall what it was

Add a reply

Find answers from the community

i'm working on a RAG app using