Three theses on what a research library is, what it is for, and why it deserves a tool that respects what you have built.
« Who are we, who is each one of us, if not a combinatoria of experiences, information, books we have read, things imagined?
Each life is an encyclopedia, a library, an inventory of objects, a series of styles, and everything can be constantly shuffled and reordered in every way conceivable.
A research library is intelligence already deposited — years of distinguishing, classifying, annotating, cross-referencing. ARCHILLES is the layer where it meets the artificial kind. Flat RAG cannot do this work, because the work was already done.
A research library is not a content blob to be re-uploaded somewhere. It lives on the scholar’s disk, in the filesystem they have lived in for a decade. Archilles indexes it where it already is — and the index sits next to it. Nothing leaves the machine unless the scholar deliberately wires up a hosted model.
Tags, highlights, margin notes, reading status, custom fields — the labour of curation is invisible from outside and almost entirely absent from the file itself. Industrial RAG ignores all of it because it cannot generalise across customers. Archilles indexes it as first-class query material. Your tag schema carries more signal than any clustering algorithm could recover, because it encodes a judgement no algorithm can make.
Every result in Archilles points to a precise location in a specific document, and refuses to summarise what it cannot point at. Hallucinations are for dreamers, not scholars. Page, chapter, section — every time, exportable to BibTeX, RIS, EndNote.
The hardest part of intelligent search is intelligence.
Scholars bring their own.