Quickstart

Nabu is operated from the command line and lives entirely under its own directory — plain files plus SQLite, no services. Initializing your own library takes five steps; the starter shelf below was measured at about 690 MB on disk and a few minutes of syncing on the reference machine (2026-07-13).

Prerequisites

• Ruby 3.3 or newer and git. macOS (Apple Silicon) is the development platform; nothing in the core is known to be Mac-specific, but no other platform is exercised.
• Disk: about 1 GB for the starter shelf (canonical files plus the derived databases). A full library build currently occupies roughly 45 GB canonical + 43 GB derived SQLite on the reference machine (2026-07-22; the Chinese, Japanese, and Germanic libraries dominate) — growth is entirely at your discretion, shelf by shelf.

1. Clone

git clone https://github.com/arvicco/nabu
cd nabu

2. Install

bundle install

The dependency set is deliberately small (thor, sequel, sqlite3, nokogiri, faraday, plus test tooling). Configuration is optional: every key in config/nabu.yml has a working default, so a fresh checkout runs as-is.

3. Sync the starter shelf

bin/nabu quickstart

This first syncs the core group automatically — the registry instruments the library itself depends on (nabu-lects, nabu-places, CIGS: a few hundred KB total, pre-enabled by nature; bin/nabu sync core re-runs just that sweep any time) — then the four curated starter sources, each through its normal fetch → load → index path, and ends by printing the first three commands to try. The starter set, with sizes measured from the live canonical tree on 2026-07-13:

On the reference machine the four first syncs took roughly three minutes of fetch-and-load combined (the lexica clone dominates at about two minutes; sblgnt takes seconds); allow up to ten minutes on an ordinary connection, including the per-source index rebuilds. The command is idempotent — re-running it is an ordinary re-sync — and one source’s failure never stops the rest: failures are reported at the end. bin/nabu quickstart --list prints the set without syncing anything.

4. First marvels

The starter shelf lights the library’s three signature surfaces.

One verse across seven witnesses. bin/nabu align "MARK 2.3" renders the same verse in Greek (twice — PROIEL and SBLGNT), Latin, Gothic, Classical Armenian, Old Church Slavonic, and Old English, each with its license label. From a live run (2026-07-11, on the reference box, whose fuller library attests further witnesses beyond the starter seven):

$ bin/nabu align "MARK 2.3"
MARK 2.3 — New Testament (parallel witnesses)
  13 of 13 witnesses attest this ref

greek-nt — The Greek New Testament [grc]   license: nc
  urn:nabu:proiel:greek-nt:6563
    καὶ ἔρχονται φέροντες πρὸς αὐτὸν παραλυτικὸν αἰρόμενον ὑπὸ τεσσάρων.

latin-nt — Jerome's Vulgate [lat]   license: nc
  urn:nabu:proiel:latin-nt:10368
    et venerunt ferentes ad eum paralyticum qui a quattuor portabatur

gothic-nt — The Gothic Bible [got]   license: nc
  urn:nabu:proiel:gothic-nt:37435
    jah qemun at imma usliþan bairandans, hafanana fram fidworim.

marianus — Codex Marianus [chu]   license: nc
  urn:nabu:proiel:marianus:36421
    Ꙇ придѫ къ немоу носѧште ослабленъ жилами. носимъ четꙑрьми.

wscp — West-Saxon Gospels [ang]   license: nc
  urn:nabu:proiel:wscp:102359
    & hi comon anne laman to him berende, þone feower men bæron.

… (Armenian, SBLGNT, the WEB English, Clementine Vulgate, and the four CCMH OCS witnesses trimmed)

Search by dictionary form. bin/nabu search --lemma λέγω finds every inflected attestation over the gold treebank annotations — λέγουσι, λέγοιεν, and the suppletive εἶπας/εἰπεῖν that no surface-string query can reach. Diacritics are optional on the query.

Open the dictionary. bin/nabu define λόγος prints the whole LSJ entry, with its citations resolved to passages in your own catalog where the corpus holds them (nabu show <urn> opens the cited line); bin/nabu define virtus does the same in Lewis & Short.

5. Connect your AI tools

The repository ships .mcp.json, so opening the directory in Claude Code registers the read-only MCP server automatically — the assistant can search, read, align, and define against everything you have synced, every passage carrying its license class, while remaining structurally unable to write to the library. Registration recipes for Claude Desktop and other clients, the tool reference, and the quoting etiquette are in docs/mcp.md.

Grow the library

Each further shelf is one nabu enable <source> (or enable <axis> for a whole desk) followed by its sync; bin/nabu sync --all re-syncs every enabled live-policy source in one pass. The full menu is config/sources.yml; real on-disk sizes for the larger shelves, measured 2026-07-26:

What each shelf is good for is surveyed on The Library; every text keeps its upstream license, recorded per document — the classes and per-source terms are on Sources. The non-commercial shelves are for research use and are never redistributed by the tooling.

The same walkthrough, kept in the repository alongside the code, is docs/quickstart.md.

Your desk

The starter shelf is running; the natural next move is to find the desk for your own field. bin/nabu list --axis prints the shelf census grouped under the 25 research desks — scholarly hats over the same sources, from the Classicist to the Assyriologist. Pick the one that fits and sync its members in a single command — bin/nabu sync celtic, or whichever axis is yours — then open its page: each desk carries its own member shelves, CLI recipes, search modes, and terminal setup for its scripts.

bin/nabu list --axis        # the shelf census, grouped by research desk
bin/nabu axis celtic        # one desk's card: members, holdings, gold coverage
bin/nabu sync celtic        # sync that desk's enabled members