Hi, we generate a table of contents (ToC) for each document, so LLM can use the ToC to navigate the long documents, which can bypass the context limit. You can checkout this notebook for a quick toturial about our method: https://docs.pageindex.ai/cookbook/vectorless-rag-pageindex