Three questions we hear, answered by how each approach works. Measured differences are on the benchmark page.
Text RAG extracts the words from each file, cuts them into chunks and searches those. It is fast and cheap, and it works well on clean prose.
It loses what is not plain text: which header a number sits under, what a slide’s chart means beside its title, anything in a screenshot, a diagram or a scan. Graph-Me-Maybe reads the page as drawn, so those stay findable — and the answer points to the exact spot.
Graph methods connect entities across documents, which helps questions that span files. Most build the graph by asking a model about every chunk of text, and keep the graph and its summaries beside the documents.
Graph-Me-Maybe builds its graph from the structure files already have — sheets, slides, headings, cells — and from what documents cite and resemble. And because it reads rendered pages, it sees what text-based graphs miss.
A chat assistant answers about the files attached to that conversation.
Graph-Me-Maybe indexes whole drive folders once, keeps how the documents relate, and shows the cell, slide or paragraph behind each answer — on your computer.