3D Codex

Manuscripts in three dimensions.

3D Codex turns the documentation of a book’s conservation into an object you can handle, and lets you read the book and question it.

3D Codex is not a project of the Vatican Library. It is a personal project that uses open data made available by the Vatican Library.

Turning the books and reading the pages is open to everyone. Ask the book needs an account, the same as for LOMO.

What it does

Borg.ar.229: a restoration in three dimensions, with the text of every page in Arabic and in English, and Ask the book.
Vat.ar.1450: a restoration in three dimensions.
Ask the book: a question in any language gets an answer drawn only from the pages read, with the folio of each passage. When the pages say nothing, the answer says so.

Source. The digitised manuscripts of the Biblioteca Apostolica Vaticana, read through IIIF, the open standard for heritage images. Images are fetched, read and never stored.
Reading. Claude Opus 5.5 transcribes each page of Borg.ar.229 from its photograph, line by line, in Arabic; Claude Haiku 5.5 translates the transcription into English. 488 pages, organised by treatise. This is a machine reading, not yet checked by a specialist.
Model. The conservation data model of the Restitutio project: CIDOC-CRM, CRMsci and a conservation extension, with SHACL shapes and SKOS vocabularies.
Answers. Pages are indexed with multilingual embeddings; for each question the closest pages are given to Claude, which must cite them. Daily and per-visitor limits keep the cost bounded.

3D: three.js (WebGL), in the browser
Images: IIIF https://digi.vatlib.it/
Data model: CIDOC-CRM, CRMsci, CONS (cons:, the extension of CIDOC-CRM designed for conservation in the Restitutio project), SHACL, SKOS
AI: Claude API: Opus 5.5 reads the pages and answers, Haiku 5.5 translates
Search: sentence-transformers (multilingual-e5-small), sqlite-vec
Application: Python, FastAPI; Docker, Coolify on an OVH server in France