I am facing issues in RouterQueryEngine, I have around 6 tools that I am passing to it, and I get some werid behaviour in which it provides right explanation of the tool that it want to select but it returns wrong index number. Due to this the RouterQueryEngine ends up choosing wrong QueryEngine to execute.

This doesn't happen reqularly but its quite frequent though. Any tips to solve this would be greatly appreciated

@Logan M - I can see strict mode enabled in OpenAI, but the same is not enable in through AzureOpenAI, is this right?

Can a workflow be converted to a tool?

Has someone created or has the knowledge of a RetrySQLQueryEngine? - the one which retries to generate proper sql statement if it errors out once?

I have a usecase, in which I need to ingest over 1000+ documents and over which I need to create a VectorIndex and SummaryIndex. I am able to sucessfully create the VectorIndex by adding metadata and retrieving using VectorIndexAutoRetriever. However, I am stuck at creating the SummaryIndex as I need to retrieve only one particular document (identified by the metadata) and create its summary. How can I achieve this?

@Logan M - I am trying to understand how can i use RouterQueryEngine with Chat History. If i just stuff the query with all the chat messages will it work? or do i need to do something else? How can I achieve this?

I would also like to point out that i tried playing with max_colwidth parameter, but it didn't make any difference. Here is how I am setting it -

How can I see the formatted prompt which is sent to the LLM? I can see what kind of prompts are used by using query_engine.get_prompts() but not the formatted ones which are sent to LLM.

Hi, how do I see the embeddings that are generated by the VectorStoreIndex for each of the nodes?
When I iterate over the nodes of the index, I find that the TextNode object has embeddings=None.
Now, I am wondering where the embeddings are stored.

I am trying to extract structure data from the unstructured text using structured_predict, but if my Pydantic model has a field with type as datetime.date, llama-index is not able to parse the string and throws an error saying "Invalid Date Format". How can this be rectified

load_index_from_storage vs VectorStoreIndex.from_vector_store() -> does both of these do the same thing?

Because when I see the print the index.docstore.docs from the index returned by load_index_from_storage , I can see the documents, but same is not the case when I do it for the index returned from VectorStoreIndex.from_vector_store()

How can I store the indices in a VectorStore such that I will be able to build a RecursiveRetriever after loading it from a persistent storage? Currently I am finding it difficult to populate the node_dict parameter while creating the instance of RecursiveRetriever

So here is the scenario that I am trying to accomplish, I have a pdf containing text + images + tables. I need to develop a RAG which is needs to retrieve images along with the text based on the query relevance.

I can achieve this by creating TextNodes, ImageNodes and IndexNode and then using RecursiveRetriever to retrieve the nodes along with the images.
However, this approach has a problem where in, if the there are more TextNodes with the relevant text (more then similarity_top_k) then the ImageNode wont be retrieved.
To avoid this, is it possible to do some workaround (or have a feature in the library) such that the RecursiveRetriever retrieves TextNodes and ImageNodes separately along with the scores, so that as a user I can decide weather to pass just the TextNodes or TextNode + ImageNode to the LLM in its context.

This use-case is an important one IMO, and I feel that this should be built in the library, I would love to hear some discussions on this and more than happy to contribute if the need araises.