Sources

Nabu is an aggregation of other people’s scholarship. Every text in the library exists because an upstream project digitized, edited, annotated, and published it under terms that permit this kind of use; those projects are credited here, and their license terms travel with every passage. The maintained inventory, including sources evaluated but not (or not yet) ingested, is docs/02-sources.md in the repository.

The licensing model

Every document carries a license class, recorded as data at ingestion and displayed on every surface — search hits, exports, alignment rows, and MCP responses alike:

• open — public domain or CC0. No restrictions.
• attribution — CC BY or CC BY-SA class: redistributable with credit. Together with open, this covers roughly 99% of documents.
• nc — CC BY-NC-SA class (GRETIL, most treebanks): licensed for non-commercial use. Appropriate for private research, including AI-assisted reading; never redistributed by the tooling.
• research_private — sources whose terms (for instance CC BY-ND, or scholarly-use grants without a redistribution clause) do not fit the classes above. Held for personal research only, and excluded from the MCP surface by default: an AI client must opt in per call to see them. The Freising Manuscripts edition (CC BY-ND 2.5 SI) is the first source behind this gate.

Two further habits belong to this model. First, licenses are read from upstream metadata wherever it is machine-readable (ORACC’s per-project license field, for example), never hardcoded, and per-document overrides record cases where one collection carries differently licensed layers. Second, texts an upstream later deletes are retained and remain searchable — honestly labelled as retired upstream — under the license they were fetched with.

This posture is deliberately conservative: the library is personal research infrastructure, not a redistribution service, so it can hold non-commercial and no-derivatives material lawfully while keeping the boundary explicit in the data. Should any part of the collection ever face outward, the license classes are the gate that decides what may pass.

Corpus sources

All 159 sources below are synced and live as of 19 September 2026, listed alphabetically by source id. The same sources grouped by research desk — each desk’s member shelves, license mix and sync status on one page — appear on the research axes.

Feature modules

Eighteen further registry rows are feature modules, not corpora: machinery that fetches reference data but mints no documents of its own.

The local shelves

Four further registered sources hold no upstream at all — they are the library’s shelves for authored and acquired material, synchronized by re-scanning local files rather than fetching. The language-dossier shelf (local-language) carries the library’s own per-language curation, one Markdown dossier per language code. The local-library shelf (local-library) files the owner’s PDFs, scans, and offprints through the nabu ingest command (which also accepts http(s) URLs, downloading first and recording the address in the manifest); everything on it defaults to the research_private class — catalogued and searchable locally, never served or redistributed. The source-dossier shelf (local-source) carries a curated description of every registered source, served on the nabu list census. The notes shelf (local-notes) records the owner’s annotations on any citable URN through nabu note. What these shelves hold is, by design, the owner’s private business — their contents and counts appear nowhere public. Licenses on these shelves belong to whatever the owner files there; the class system above is the gate that keeps restricted personal material private.

Acknowledgements

This library would be empty without the sustained, mostly under-funded work of the projects above: the Perseus Digital Library and Open Greek & Latin; the Duke Databank and papyri.info; the ORACC consortium and its constituent projects; the PROIEL, TOROT, and ISWOC treebank teams in Oslo and Tromsø; the Universal Dependencies community; GRETIL at Göttingen; the Corpus Cyrillo-Methodianum Helsingiense and Kielipankki; CLARIN.SI and LINDAT/CLARIAH-CZ; the Bosworth-Toller digitization at Charles University; the eZISS edition of the Freising Manuscripts; the Society of Biblical Literature and Logos; eBible.org and the open-bibles collection; the Oxford Text Archive; the Wiktionary community and the Wiktextract project; the Cologne Digital Sanskrit Lexicon; Coptic Scriptorium; the Epigraphic Database Heidelberg; the IE-CoR project; the LiLa (Linking Latin) / CIRCSE group, whose linked-data editions carry LIV and de Vaan; George Starostin and the StarLing / Tower of Babel project, with the compilers of its constituent databases; ZRC SAZU, whose dictionaries carry the Slovenian lexicographic tradition; the ChronHib project at Maynooth (CorPH); the RIIG project at Ausonius / Bordeaux; the Ogham in 3D project at DIAS; the Hethitologie-Portal Mainz and the TLHdig team; the Cuneiform Digital Library Initiative; the electronic Babylonian Library at LMU Munich; the ETCBC at the Vrije Universiteit Amsterdam, whose text-fabric editions carry the BHSA, the Scrolls and the Peshitta; the CACCHT project in Copenhagen; the Srophé / Digital Syriac Corpus editors; the ETCSL project at Oxford with the Oxford Text Archive; the Kanseki Repository (Christian Wittern) and CBETA; the Thesaurus Linguae Sericae editors; the Unicode Consortium’s Unihan database; the nk2028 group’s Qieyun-system edition; William Baxter and Laurent Sagart; the HDIC project; the Electronic Dictionary Research and Development Group (James Breen); Andrew West’s BabelStone; the Oxford-NINJAL Corpus of Old Japanese; the volunteer transcribers of Aozora Bunko; the Menotec project and the CLARINO INESS infrastructure at Bergen; George Walkden’s HeliPaD; the Referenzkorpus Mittelhochdeutsch teams at Bochum and Bonn; and Uppsala University’s Samnordisk runtextdatabas — this library is based on the Scandinavian Runic-text Database, as its Open Database License asks to be said. Users of this software are bound by, and should credit, these upstream projects under their respective terms.

Sources that could not be ingested for license reasons — however valuable — are recorded honestly in the repository inventory with the specific blocking terms and possible unlock paths (for instance TITUS, whose scholarly-use terms grant no redistribution, and the Rahlfs Septuagint under CATSS conditions).