Does it support large-scale multimodal documents?

[wangxingjun778]

This is a very interesting approach with a high degree of completeness. 😄👍🏻

Additionally, I’m curious whether there are challenges when dealing with large-scale documents—such as high token consumption, slow tree structure generation, retrieval efficiency issues, multi-hop reasoning problems, or difficulties in understanding multimodal content.

[tmishinev]

I was looking at the same issue. Looks good for small collections but for large datasets you will need some tree selection before the retrieval step I guess. There is some hint that there will be a follow up:

Multi-Node Reasoning with Content Extraction — Scale tree search to extract and select relevant content from multiple nodes.
Multi-Document Search — Enable reasoning-based navigation across large document collections, extending beyond a single file.
Efficient Tree Search — Improve tree search efficiency for long documents with a large number of nodes.
Expert Knowledge Integration and Preference Alignment — Incorporate user preferences or expert insights by adding knowledge directly into the LLM tree search, without the need for fine-tuning.

[denis-samatov]

On the scale question @tmishinev's roadmap list is accurate as of when it was posted, and it's since shipped in part: cross-document reasoning over large corpora is now available via PageIndex Cloud's File System (Cloud-only, tree-indexes an entire corpus rather than one document) — see https://pageindex.ai/blog/pageindex-filesystem. The open-source/local path is still single-document, so for large-scale multi-document retrieval today you'd reach for Cloud rather than rolling your own tree-selection layer. I don't have confirmed specifics on multimodal-content understanding beyond what's already in this thread, so leaving that part open.