Find answers from the community

Updated 3 months ago

If I understand correctly a full "

If I understand correctly a full "llamaindex index" is made of :

one vectorStore
one docStore
one indexStore

I understand the need for the vectorStore. What is the usage of the 2 others, especially for VectorStores that actually store text ? Are they mandatory or not always needed ?

5 comments

LLogan M

the other two are not mandatory. Especially when using a remote vector db

AArcy

it does not seem to work without docStore for automerging retriever :

automerging_retriever = AutoMergingRetriever(retriever, context, verbose=True)
auto_merging_engine = RetrieverQueryEngine.from_args(automerging_retriever, llm=self._llm.model)
retrieved_nodes = auto_merging_engine.retrieve(query)

LLogan M

yea some featuers require a docstore, youd have to maintain and persist one

LLogan M

docstore.add_documents(nodes)

LLogan M

its not too bad

Add a reply