Mouseion OS (MOS)

How It Started

I had just downloaded Christopher Andrew's thousand-page history of MI5. James Gleick's The Information was on the shelf. And somewhere on my server, a pipeline had been crawling World War II dossiers for a few weeks, looking for names, dates, and contradictions.

Three sources. Three completely different worlds. But in my head, they kept crossing.

Claude Shannon — the father of information theory, and the man Anthropic's Claude is named after —, lived in the same era as the MI5 officers in Andrew's book. Wartime intelligence services used the same principles of coding and noise that Shannon would describe mathematically. And my research on the Life Lens System was trying to understand how the same reality looks different depending on which lens you use.

I thought: what if I don't just read all these sources, but let them read each other?

I named it Mouseion, after the Mouseion of Alexandria — the research institute attached to the famous library. It wasn't the storage that was special. It was that people sat there who read the books and crossed them.

Put a book into the system, and the system reads it. It extracts all persons, all organisations, all dates, all source references, all concepts. Then it crosses them with everything already in the database.

Shannon appears in Gleick. But also in Bateson. Bell Labs shows up in three sources that seem to have nothing to do with each other. The word "information" means something different in Shannon (a mathematical quantity) than in MI5 (intelligence) than in Husserl (intentional consciousness). That difference becomes visible.

Each component exists as a standalone tool or academic concept. Microsoft has GraphRAG for entity extraction. Stanford researches semantic change over time. The platform nodegoat builds timelines for historical research. But nobody has combined them into a personal research instrument that treats books as archives, traces concept migration across domains, and builds its own lexicon from the structure that authors have already provided.

The building blocks exist. The building doesn't. Until now.

Four Halls and a Pneumatic Post

The original Mouseion of Alexandria had a peripatos (covered walkway), an exedra (discussion hall), an oikos (dining hall), and the famous library. MOS has four halls and a connection layer that runs through all of them — the Pinakes Pneumatic Post, named after Callimachus' catalogue, the first library classification system in the Western world.

Five Roles

The historical Mouseion had scholars with distinct functions. MOS has five roles, each named after an Alexandrian tradition:

The AI librarian of my Mouseion is called Cal — named after Callimachus of Cyrene (ca. 310–240 BCE), the scholar who built the Pinakes, the first library catalogue in the Western world. Callimachus wasn't the head librarian — that was Zenodotus. He was something more interesting: the man who decided how to organise everything. He invented classification by genre, alphabetical ordering by author, and the idea that every entry should carry metadata: title, birthplace, teacher, number of lines, opening words. Twenty-three centuries before Dublin Core, Callimachus wrote the spec.

Where Tom is my guide in the Life Lens System — built on 560,000 records from my own life — Cal is his counterpart for MOS: a librarian who doesn't just catalogue, but also connects sources to each other, and works in close contact with Tom. When a new source enters the system, Cal decides where it goes, what gets extracted, and how it connects to what's already there.

Three Thinking Tools from Alexandria

The Alexandrian scholars didn't just store knowledge. They invented methods for thinking about it that are still in use 2,300 years later. Three of them are fully integrated into how MOS works:

Obelos — "How certain am I of this?"
Zenodotus placed a small horizontal stroke next to lines in Homer that he suspected were not authentic. He didn't delete them — he marked them. That single gesture invented source criticism. In MOS, the obelos became twelve certainty layers. Every claim carries its doubt. Nothing is destroyed, everything is marked.

Scholion — "What did I think when I read this?"
Alexandrian scholars wrote comments in the margins, never in the text itself. The thinking about the source stayed separate from the source. The text is sacred; the margin is yours. In MOS, the scholion became the Exedra — the reasoning log where turning points, doubts, and connections are captured. The thought that crosses your mind while reading chapter seven may be more valuable than chapter seven itself.

Ploi — "What exists out there that I don't have yet?"
The Ptolemies had a harbour policy: every ship entering Alexandria had its books confiscated. The library copied them, then returned the originals — sometimes. In MOS, Ploi is the acquisition layer: actively looking outward, bringing knowledge in. I gave Cal two library cards — one for the Dutch public library, one for the Internet Archive. Cal borrows books, picks them up, reads them, and returns them. The harbour of Alexandria is now a municipal library card and the largest digital archive in the world.

Anagnostes — the reader

In the ancient Mouseion of Alexandria, the Anagnostes (ἀναγνώστης) was the reader — the one who made texts accessible to scholars. Not just reading aloud, but interpreting, connecting, revealing what was hidden in the scrolls.

Ana does the same thing, at scale.

Most digitisation tools stop at recognition: turning a scan into text. Ana reads further. It identifies who appears in a document, what claims are made, which concepts are at play — and where sources cross-reference each other.

A scanned archive is a collection of images. After Ana, it's a searchable network of people, organisations, claims, and connections. Patterns emerge that no human reader would find across thousands of pages.

Ana processes source material through OCR and HTR, extracts entities and concepts, and makes collections cross-searchable. Every source — books, articles, dossiers, reports, blogs, clippings, and even songs — passes through the same steps:

What gets extracted, why, and how

Persons with role and context — not just "Shannon" but "Claude Shannon, mathematician, Bell Labs, information theory". Why: a name without context is useless — knowing someone is a physicist changes how you read their claims. How: the system reads each passage and identifies who appears, in what role, at what organisation. Certainty: 95%.

Concepts with meaning — what the author means by "entropy", "intelligence", "information". Why: the same word means different things in different fields — tracking that is how you find real connections and avoid false ones. How: extracted per chapter, then traced across domains. Certainty: 90%.

Claims by the author — assertions, not summaries. What the author states as true. Why: a summary loses the author's voice. A claim preserves exactly what they committed to — and can be crossed with claims from other authors. How: each passage is scanned for declarative statements that the author presents as fact. Certainty: 90%.

Cited works — title, author, year. The citation network, extracted from non-academic sources. Why: who cites whom reveals intellectual lineage — which ideas flow where, and which authors are bridges between fields. How: bibliographies, footnotes, and in-text references are parsed automatically. Certainty: 99%.

Twelve certainty layers

Every piece of knowledge in MOS carries a certainty layer, from 1 (mathematically proven) to 12 (unknown / open question). The Critic assigns the layer. The Fact-Gate decides whether to store, verify, or park in the Exedra.

Shelf-Keying

The most interesting discovery was that you don't need artificial intelligence to read a book. A good non-fiction book reads itself.

How? A book has an index — literally a list of all important persons, places, and concepts, with page numbers. The author already did the work. The bibliography is the citation network, ready-made. The table of contents tells you the structure. Footnotes mark the source references.

We called this Shelf-Keying: every book you add makes every other book smarter, because the keys from book A are also tried on book B.

The pun is free: shelf sounds almost like self. The shelf that reads itself.

How do you know the "Shannon" in book A is the same as "C.E. Shannon" in book B? Not just by name, but by company. If Shannon appears next to "entropy", "Bell Labs" and "bit", it's the mathematician. If Shannon appears next to "river", "Ireland" and "fishing", it's the river.

Every entity builds a fingerprint: the words that appear in its vicinity. With every book, the fingerprint sharpens. After five books, a false match is nearly impossible.

Named After Alexandria

The Mouseion (Greek: "seat of the Muses") was a royal research institution in Alexandria, founded around 280 BCE by Ptolemy I Soter. Not a museum in the modern sense — a living research community, comparable to the Institute for Advanced Study in Princeton.

30 to 50 scholars resided there. Salary from the state, tax-exempt. No teaching obligation — free research. Common property, shared meals. Disciplines: mathematics (Euclid), astronomy, medicine (Herophilus performed the first dissections), geography (Eratosthenes calculated the circumference of the Earth), literary criticism, mechanics.

Callimachus (ca. 310–240 BCE) created the Pinakes: 120 book rolls, the first library catalogue of the Western world. Classification: class, subgenre, alphabetical by author. Per entry: title, author, birthplace, father, education, number of lines, opening words. The model dominated for centuries.

The Pinakes are not a fifth hall but the pneumatic post between the four halls. Not an address — the connection that runs through everything.

Every concept in MOS maps to an Alexandrian tradition. The obelos (the critical mark) became our certainty layers. The scholion (margin note) became our turning point capture. Ploi (the harbour practice of confiscating books from ships to copy them) became our acquisition layer.

What It Found

Christopher Andrew's thousand-page MI5 history was the first book through the pipeline. The system extracted thousands of entities, built a timeline spanning 1909 to 2009, and began crossing them with WO2 dossiers that were already in my database.

The crosses revealed things that neither source could show alone. Names that appeared in MI5 files but were absent from the Dutch standard work on World War II. Dates that contradicted across sources — not errors, but research questions. Patterns of omission that became visible only when you could search across everything at once.

My WO-II-OS archive — applying the Life Lens System to World War II — now contains 2 million records, 1 million unique persons, and 196,000 confirmed by two or more independent sources. The Groninger Oorlogspuzzels are a publicly visible part of it. MOS is the reading layer underneath: it ingests the archives, extracts the entities, and feeds them into the cross-referencing engine.

Seven findings that don't appear in the secondary literature. Not because nobody looked, but because nobody could search across all sources simultaneously. The details are documented but not yet public.

Part of the Life Lens System

Mouseion OS is the library inside my Life Lens System — the memex I built. MOS is one of many tools that strengthen and extend my LLS, the way a library strengthens a university: not the building itself, but indispensable for everything that happens inside it.

Tom looks inward — at my life. Cal looks outward — at knowledge about the world. Together they form a complete system: personal data crossed with source material, lived experience crossed with what others have written.

MOS is the layer before thinking. ThetaOS is the layer after.

Journal

Discoveries, methods, and lessons from building a self-reading library.

Timeline

Changelog

Born from the Pilot Informatie Autonomie

This project is a by-product of the Pilot Informatie Autonomie, where we discovered that to work with computers, information, and especially AI, file formats that require software to open them — DOCX, PDF, ODF — are a dead end.

Everything you see here works because we use formats that computers can read, search, and process natively: plain text, in formats like .md, .json and .txt.