A book that grows.

When I started reading about Mesopotamia, I kept hitting the same wall: the material is scattered, and nobody had pulled it together into one continuous, readable story. The Cuneiform Digital Library Initiative (CDLI) holds the catalogue of 340 000 tablets, but as a scholar's search engine. The Open Richly Annotated Cuneiform Corpus (ORACC) holds tens of thousands of scholarly editions split across thirty sub-projects, each its own portal. The Electronic Text Corpus of Sumerian Literature (ETCSL) is frozen since 2006. The State Archives of Assyria online (SAAo) covers one period and one language. The royal-inscription series (RIME, RIAo, RINAP) each tell one slice of one king-list. The Met Museum's Heilbrunn Timeline gives a beautiful chronological narrative — but in four short essays, without the tablets. The big museums (British Museum, Penn, Yale, Louvre, Berlin) each have their own collection portal, never linked to the others.

On the AI side, a few projects translate cuneiform with machine learning. The closest in spirit is AICC (aicuneiform.com), built by Frank Krueger in 2023 — a fine-tuned T5 model trained on CDLI and ORACC, which has put roughly 130 000 AI-translated tablets online. It is searchable by period, genre, object type. But it is a flat database, frozen, with no narrative, no chapters, no plain- English interpretation, and no daily improvement.

So, no site I could find combined all of the following: aggregating CDLI + ORACC + ETCSL + SAAo + RIME/RIAo/RINAP + Wikimedia in one place; presenting them as a chronological book with prologue, twelve period chapters, and an epilogue; using a frontier large language model (Claude) to translate the tablets that scholars have not yet translated; writing a plain-English interpretation for each one so a curious general reader can grasp what they are looking at; and running a daily self-improvement loop where the engine's glossary grows, older translations are refreshed when the prompt is upgraded, and dates are refined from year-name evidence in the text itself.

That is the gap this site is trying to fill — humbly, in public, one tablet at a time.

Humanity's first great civilization left us hundreds of thousands of clay tablets — administrative records, royal inscriptions, hymns, legal codes, mathematical tables, love letters, school exercises, flood myths. They are the rawest evidence we have of where we came from. For the non-specialist reader, they remain almost inaccessible: the catalogues are technical, the languages unfamiliar, the translations dispersed across decades of philological literature.

This project weaves them together. Period by period, theme by theme, we translate, annotate, and integrate the tablets into a single living text — a book the public can read, with every claim traceable back to its primary source.

For each tablet, we present four layers visible side by side:

• The image — photograph or stylized representation, with full attribution.
• The transliteration — the cuneiform signs rendered in Latin script, drawn from established corpora (CDLI, ORACC).
• The translation — when a scholarly translation exists in open form, we use it. When it does not, we propose one with AI assistance, always with a confidence indicator visible. See the engine for how this works.
• The note — a short scholarly contextualization, placing the tablet in its time and significance.

Every translation labeled experimental is an AI-assisted hypothesis, not a substitute for trained Assyriological work. We do this transparently: experimental status is a feature, not a flaw, and scholar-validated corrections are welcome.

The project builds on open scholarship. Tablet metadata, transliterations, and imagery come from a careful selection of compatibly licensed sources:

• CDLI (Cuneiform Digital Library Initiative) — CC-BY
• ORACC (Open Richly Annotated Cuneiform Corpus) — mostly CC-BY-SA
• The Met Open Access — CC0 for many tablet objects
• Yale Babylonian Collection — open for research
• Wikimedia Commons — for representative imagery of iconic public-domain artifacts
• Public-domain translations (Kramer, etc.) and our own annotations

Every item carries its license and attribution. Sources requiring permission (ETCSL re-use, British Museum imagery, etc.) are linked rather than copied, or used only under explicit agreement.

Done. Twelve period chapters from Ubaid (5500 BCE) to Achaemenid Persia (330 BCE), plus a prologue on the Neolithic gestation that preceded the Sumerians and an epilogue on the long silence that followed and on the nineteenth-century rediscovery of cuneiform. Over 106 000 tablets ingested from CDLI, ORACC, ETCSL, SAAo, RIME / RIAo / RINAP and Wikimedia, with ~96 000 photographs cached locally. A six-theme thematic index. A two-lane chronological frieze on the home page (Assyria north, Babylonia south). A live stats panel measuring corpus / translations / photos as they grow.

Live now. An automated daily loop runs at 03:15 server time: translate-N-new + refine-N-dates + harvest glossary + refresh-M-stale + rebuild index + snapshot stats. The engine improves with every batch — its glossary grows from the previous run's uncertain terms, older translations are upgraded when the prompt is bumped, and tablets shift in the chronology when the engine extracts year-name formulas from the text. See /engine.

Coming. Scholar validation workflow, an open API for researchers, sign-by-sign tooltip glossary in the transliteration view, a second-model critic pass on engine translations, multilingual interface starting with French. As the engine grows more confident on a tablet, its confidence badge climbs from experimental to medium to high. The work is open-ended.

For corrections, scholarly collaboration, photographic donations, or help funding the AI inference: please use the contact page. See also privacy and terms.