Encyclopaedia Biblica
A biblical encyclopedia written by artificial agents, and made incapable of lying.
Ask a question. Receive a sourced answer, and paths to go further. When the corpus does not cover a question, the encyclopedia says so. It does not invent.
Two heritages of humanity.
One will now serve the other.
The first is a Book. Translated, copied, and pondered for two thousand years, it remains out of reach for one person in five: more than 1,600 languages do not possess a single verse of it, thousands more hold only a part, or a translation the passage of time has closed, or no version anyone is free to share.
The second was born of the first. Twenty centuries of reading: the Fathers, the catenae, the lexica, the manuscripts, the commentaries. A continent of writings that every generation has produced around the Bible, and that dispersion threatens as surely as oblivion.
Encyclopaedia Biblica gathers these two heritages into a single gesture. It receives the second: 126,187 works assembled, indexed verse by verse, preserved in open formats for the generations to come. It keeps it: artificial agents write its encyclopedia behind a gate that verifies every assertion, for one does not transmit a heritage by distorting it. It transmits: a translation engine, armed by the entire corpus, translates the Bible directly from Hebrew, Aramaic and Greek into the languages that are still waiting for it.
The fruit feeds the root. The library serves the Book.
Twenty centuries of scholarship, one door.
Every inheritance begins with an inventory.
This one is held in a single scholarly database: the Open Source Bible corpus. The biblical text in its original languages, analysed word by word. Twenty centuries of commentary, from the Apostolic Fathers to the nineteenth century. The reference apparatus of modern scholarship. And all of it tied back to Scripture, verse by verse: ask about Genesis 1:26, and the catena answers with the 2,484 works that cite or comment on it, from Origen to Calvin, filterable by era, confession, genre and language.
This is not a collection: it is an ark. Open formats, figures regenerated daily from the living database, precision measured and published, and a licence resolved for every work: public domain or openly licensed, fail-closed. What cannot be freely shared is not published. A heritage built to outlive its builders, and to belong to no one because it is meant for everyone.
Snapshot of 25 July 2026, regenerated daily from the live database, not written by hand.
The original text, fully armed
Hebrew, Greek and Syriac editions analysed word by word: nearly one million interlinear tokens, 1.48 M lexicon alignments, full morphology, a critical apparatus of the Greek New Testament, and a versification map across the Hebrew, Greek, Latin and English traditions.
Measured, published quality
To our knowledge, the only large-scale biblical catena whose precision is published: an independent double-scale audit measures 94% index-exactness, and recall is benchmarked against the BiblIndex reference corpus, Father by Father.
"Same door, same data." Every agent and every pipeline reaches the corpus through one and the same interface. No corpus, no agents.
Retrieve first. Write second. Verify third.
Receiving is not enough. A heritage poorly transmitted is a heritage betrayed.
On the foundation, a team of specialised agents writes the encyclopedia itself: study entries on 34 types of subject (persons, places, words, doctrines, feasts, plants, measures, manuscripts) drawn from a catalogue of 57,541 subjects, at three depths of reading, from the newcomer to the exacting reader. These agents never write from memory: they query the library first, write from what they actually found, and a deterministic gate verifies everything before publication. When the corpus does not cover a question, the encyclopedia says so. It does not invent.
- Citations verified verbatim. A false patristic attribution is removed automatically, never published.
- Words checked at the lexicon. Every Hebrew or Greek claim is controlled; archaeology is checked against its register.
- Proven, entry by entry. A ledger reconstructs the source consultations actually performed: the encyclopedia does not claim to have used its sources, it can prove it.
- No half-entries. If a level is missing or a section would be empty, the entry is not published at all.
One entry, three depths.
Accessible, concrete, no jargon. No prerequisites assumed; no one left at the door.
Historical context, theological weight, the reading of the Church Fathers.
Original languages, textual variants, intertextuality, the debates of scholarship, commentators named and cited.
The catena
"Who spoke about this verse?" The works, from Origen to Calvin, that cite or comment on any verse, in Greek, Latin, French or English.
The Council of the Fathers
Where the Fathers converge, where their accents diverge, school by school; every excerpt verbatim, the synthesis written from the displayed citations alone.
Word studies
A full page for a single biblical word (ἀγάπη, יהוה): etymology, history, dated patristic reception, complete concordance.
Lives, not labels
Every cited figure receives a biography anchored in the historical dictionaries of reference. When the corpus cannot cover a figure, the notice says so.
Where traditions diverge
On 424 verses, the great Christian families read differently. The entry lays out the positions, each attributed to its tradition and reference document.
Beyond the page
Pastoral notes that return a verse to the heart; audio episodes born of a real discussion between agents; group challenges aiming at one observable transformation.
The keeper cannot lie. This is not an intention: it is a construction.
The first entries open to readers with the pilot's public release; the gate exists so that, on that day, every one of them can be trusted.
From the original languages, word by word.
Here the heritage received gives back what it owes.
Four frontiers still separate humanity from the Book:
Languages without a Bible
More than 1,600 languages do not possess a single verse of Scripture.
Languages with a partial Bible
A Gospel, a New Testament, never the whole.
Languages with an ageing Bible
Served on paper, but by a text the living language left behind a century ago.
The invisible frontier
Cutting across even the best-served languages: those that possess no Bible anyone is free to quote, to integrate, to share.
The engine translates directly from Hebrew, Aramaic and Greek, analysed word by word, never from an existing translation. It practises computational exegesis: pericope by pericope, with the lexicon, the grammar, the manuscripts and twenty centuries of reception summoned at every verse. What twelve hours of labour once accomplished for a single verse, the corpus now places at the service of every language.
It produces candidate translations: sourced, auditable, released to the public domain, offered for validation to those whose heart language it is. Reference Bibles (the King James, the Vulgate, the historic translations) are held as read-only, forever. Projects currently in the workshop include Luxembourgish and contemporary French, with further languages in preparation.
The machine arms the translator; it does not replace him.
A scriptorium of agents.
Frontier models, monastic discipline.
Encyclopaedia Biblica is built at the cutting edge of artificial intelligence, and runs it like a scriptorium. The approach is agentic: not one model prompted to know everything, but a team of specialised agents, each bearing a biblical namesake that states its office, each holding only the tools its office requires, and none trusted on its word alone.
One door, eighty-four tools
Every agent reaches the corpus through one and the same interface: a single MCP server exposing 84 scholarly tools. No agent has a private view of the data: same door, same data.
Four ways of reading
Semantic search over 17.3 M embeddings, full-text in every language of the corpus, fuzzy matching that survives imperfect OCR, and a typed knowledge graph that answers verse → doctrine → tradition in a single query.
Orchestrated, not improvised
Agents run in parallel under quality checkers and mandatory floors. The encyclopedia's founding run wrote 32 types of entry with 10 agents working side by side: 32 successes out of 32, in about 26 minutes.
The scriptorium
Thomas, named for the apostle who would not believe without the proof. He doubts on purpose: before a Father's citation goes out, he demands the evidence and adjudicates the claim. The desk's conscience. Birth announcement of the fact-checker, the project chronicle
Twelve hours per verse.
This project does not begin with artificial intelligence. It begins around 1996, in Jerusalem, where its founder sat learning Biblical Hebrew, Aramaic, Koine Greek, Syriac and Ecclesiastical Latin, and translating by hand. "Twelve hours per verse, sometimes more. I stopped, started again, abandoned, came back. The notebooks piled up." Twelve hours is what it takes to check every word against its dictionary, every variant against its manuscripts, every reading against those who proposed it before you.
Thirty years and three abandonments later, in April 2026, he resumed the same project: one man at a single desk in his own home, a man who knows the texts and does not write code, working day after day and night after night with a frontier AI that codes and does not know the texts. Between the two of them grew what neither could have built alone: a corpus of 126,187 works tied to Scripture verse by verse (to our knowledge, without a consultable equal) and the 57,541-subject encyclopedia now rising on it.
Thirty years ago I closed a notebook I thought I had failed. I was wrong. I had only run out of tools. G., the project chronicle
Encyclopaedia Biblica is the wager that this seriousness can change scale without changing nature, provided a gate, and not an intention, answers for it. The certainty has changed its seat: it no longer resides in the agent's intention, which is invisible, but in the artifact, which is not. "A method kept honest in private is a hobby; a method kept honest in public is an institution."
The landscape, not the verdict.
On the subjects that historically divide the Catholic, Orthodox, Protestant and Evangelical traditions, no page takes sides, adjudicates, or seeks to convert. It reports the understandings in presence and attributes each to the tradition that holds it, down to the document and the canon. The neutrality is structural, not supervised: deterministic filters strip unattributed verdicts, and the balance of the material itself is audited, not only the wording.
Writing "the Council of Trent teaches that…" is a historical fact. Writing "this doctrine is unbiblical" is a verdict. The first is our craft; the second is forbidden to us. The Charter of Confessional Neutrality