Hello everyone, i am trying to store my

A summary index does not use embeddings, so there are no vector to store (and hence no vector store)

that is new information for me, if it did not use embeddings what does it used? and do you have any suggestion on how to save the summary index because i am making an agent that make use of both VectorIndex and SummaryIndex

The summary index sends every node in the index to the LLM -- this is useful for things like summaries 🙂

You need to either save to disk, or use a remote docstore+index_store like redis, mongodb

ahhh i see so i need an index store rather than vector store for SummaryIndex is that right? and also the summary index does not use neither llm or embeddings in order to make an index?

correct 👍

Thank you for the feedback!

You could write summaries as metadata in your nodes that are then stored in a vector db if you wanted @Hardrion

that would be a great idea, but how do i load those indices into SummaryIndex or do i just use the load_index_from_storage? @sysfor

and also based on the SubDocSummaryPack each node will be index, so i can just insert the index node as metadata for each node, am i right?

the latter

you would insert the parent_node summary into the meta data for each child node
child_node.metadata["context_summary"] = str(parent_summary).lstrip().lower()

Thank you, i will try this