Clarification on PageIndex Architecture: Usage of Chunking and LLMs

[aag21]

While exploring PageIndex, I understand that it aims to preserve document structure and hierarchy.

I would like to clarify whether it is possible to generate the document tree structure entirely without any LLM involvement.

Specifically:

1. Can PageIndex generate a document tree structure without using LLMs?
2. Are any LLM calls required during tree/index generation?
3. If LLMs are involved, is there a configuration or mode to disable them and still obtain the document hierarchy?
4. Which components of the tree generation process depend on LLMs, if any?

Any clarification on the architecture or implementation details would be appreciated.