Arabic — The Arabist

The Arabist — the Islamicate library whole, Quran and hadith to falsafa and adab.

The OpenITI lane: premodern Arabic and Persian literature at corpus scale — Quran and hadith, history and biography, law and falsafa, the dīwāns and adab — with the Persian shelf (Ḥāfiẓ, Ibn Sīnā) riding the same Arabic-script fold that makes ara/fas cross-searchable (P41-3) — and DiCCAS, disaster accounts excerpted from ten classical sources with catastrophe terminology tagged — and, since P96, the KITAB text-reuse instrument: upstream-computed reuse edges joining the held OpenITI books pairwise on the intertext desk.

The premodern Islamicate library is LIVE: OpenITI’s ~9,079 primary texts (~1.12 B words, the 2026 release zip md5-pinned into canonical) made Classical Arabic the largest language on the shelves — 33.3 million passages — with the Persian lane (fas) riding the same corpus. Holdings below are read live.

New here? The Quickstart sets up the library in minutes.

The shelves

A source wears every desk it serves — these four answer this desk. Holdings are read live from the catalog and dated; a shelf with nothing synced yet says so.

Languages on this desk (live doc-or-entry counts as of 19 September 2026): ara 8,738 · fas 351.

The desk’s instruments

• No gold lemmas — OpenITI is unannotated. The corpus carries no morphology or lemma layer, so --lemma, vocab and formulas do not apply to this desk; its instruments are full-text search across the whole Islamicate shelf and the timeline.
• The ara/fas Arabic-script fold (P41-3): one search skeleton across the ی/ي and ک/ك keyboard split, maqsura, taa marbuta, tashkeel, tatweel and ZWNJ — so a query typed on either keyboard reaches the stored form whichever keyboard wrote it. Search-side only; the stored bytes stay pristine (conventions §9), and --lang ara / --lang fas scope to one shelf.
• The AH death-year timeline: every OpenITI urn opens with the author’s 4-digit hijrī death year, so the OpenitiDates extractor lands each text on the calendar as a CE terminus — round(AH × 0.970225 + 621.57), the tabular conversion — and --from/--to and --century scope the shelf by when its authors died (Ḥāfiẓ d. AH 792 = 1390 CE). No gazetteer, so no --place on this desk.
• License posture — nc (CC BY-NC-SA 4.0, the Zenodo record’s only grant): the shelf is MCP-excluded, so the AI server never serves OpenITI passages; the CLI reads them for local research.

Working the arabic desk

The generic axis surfaces — every desk answers to these, in working order (enable once, sync, then query):

nabu enable arabic               # first time: put this desk's shelves in this box's profile
nabu sync arabic                 # fetch/refresh the desk's enabled members
nabu list --axis arabic          # the shelf census, this desk only
nabu axis arabic                 # the desk card: members, holdings, gold coverage
nabu search WORD --axis arabic   # a query scoped to this desk's shelves

This desk’s own surfaces:

nabu show urn:nabu:openiti:0792Hafiz.Muntasab.PDL00074-per1  # Ḥāfiẓ's Muntasab — Persian verse in the %~% hemistich notation
nabu search الله --lang ara           # Allāh across the Arabic hadith, dīwān and falsafa shelves
nabu search دانی --lang fas           # the cross-keyboard ی/ي fold (P41-3) — an Arabic-yeh query (U+064A) still finds Ḥāfiẓ's farsi-yeh دانی (U+06CC)
nabu search --from 1300 --to 1400 --axis arabic  # term-less browse of the 14th-c.-CE shelf on the AH death-year timeline (live-verified 23 July 2026)

Ask your model

With the MCP server connected, this desk answers conversational research questions. Each example ran live against this library:

• “Who reused this passage of Ibn Sīnā’s Šifāʾ?” → nabu_links urn:nabu:openiti:0428IbnSina.ShifaAfcalWaInficalat.ALCorpus00001-ara2:1.223.4 — kind=”reuse” edges from the KITAB pairwise data: the passage reappears in Faḫr al-Dīn al-Rāzī’s Mulaḫḫaṣ (two centuries later) with character/word offsets riding the edge detail — upstream- computed alignments, distinct from nabu’s own intertext detection.

Terminal setup

• Arabic and Persian (ara/fas) are RTL and reuse the hbo/arc machinery: isolates: true wrapping, the same modes, the same honesty footer. The terminal owns the direction — the iTerm2 ≥ 3.6.0 RTL toggle (Settings → General → Experimental; Terminal.app has no bidi at all).
• Shaping stays degraded even with bidi on. Arabic is a connected script and a cell-grid terminal cannot fully join it, so what you get is right-to-left, legible, unligatured Arabic — fine for scanning search hits and citations, not for sustained reading (use nabu export and a real text view for that). Font: font-noto-naskh-arabic — naskh stays legible at terminal sizes; a dedicated iTerm2 reading profile with it boosted a few points is the workable setup (docs/display.md §2).
• Nothing to strip, and no --display translit. The P41-g census found consonantal standard-block text only (no tashkeel), so default, plain and full render the same bytes (isolates aside); and Arabic romanization is deliberately not built — the standards conflict and unpointed text lacks the vowels a romanization needs, so ara/fas pass through the translit mode unchanged (docs/display.md §1d).

The full guidance, per script, is on the display page.

One of the twenty-five research desks; the flat shelf map is The Library and the reasoning is docs/axes.md.