TL;DR
A framework that uses LLMs to compile documents into a wiki knowledge base and perform reasoning-based retrieval without a vector DB.
Key features
Document Compilation: Automatically transforms documents from various formats (PDF, Word, Markdown) into summaries, concept pages, and entity pages.
Knowledge Accumulation: Unlike traditional RAG, it compiles knowledge once, maintains it persistently, and automatically generates cross-references.
Vectorless Retrieval: Leverages PageIndex tree indexing for accurate, context-aware retrieval even in long documents.
Multi-modality Support: Searches and understands not just text, but also figures, tables, and images.
Wiki Output: Generates the knowledge base as Markdown files compatible with Obsidian.
When to use it
When you need to systematically organize and manage knowledge from a large volume of documents.
When you want to accumulate knowledge instead of performing repetitive searches like in traditional RAG.
When you need to extract accurate information from long documents or multi-modal data.