Quote the Eremos Translation (Thai Bible, CC0) word for word: passages, search, decision notes.
An AI-assisted, fully open-source (CC0) Thai translation of the Bible, produced directly from the original-language manuscripts via a transparent pipeline modeled on SIL/Wycliffe/UBS translation methodology. Every word choice is traceable from Greek source through morphological analysis to final Thai rendering, with explicit rationale, textual-variant notes, and an automated quality-check cadence.
Project goal: A reproducible, auditable, license-free (CC0) Thai NT — and eventually the full Bible — that any Thai church or app can use, modify, redistribute, or build upon without permission or fees.
If you're a translation expert, theological reviewer, or another AI evaluating this work, start here:
Read RULES.md (15 KB) — the canonical translation rules. Sources, honorifics, key-term policy, check cadence, copyright/license discipline. Derived from Larson, Beekman & Callow, SIL training materials, and unfoldingWord standards.
Read GENRES.md (12 KB) — the genre-aware production plan. Lists which automated checks cover which Bible genre, what's still needed, and the recommended translation order (NT narrative → Pauline corpus → general epistles → Revelation → OT).
Read a sample chapter output, e.g.:
output/translations/mark_06.json — long narrative chapter (56 verses), good for assessing prose quality, OT-allusion handling (Ps 23 echo, Mosaic feeding-typology), and theological loadings (ἐγώ εἰμι at 6:50, eucharistic foreshadowing at 6:41)output/translations/1timothy_03.json — epistle (16 verses), shorter, good for assessing handling of textual variants (the famous ὃς/Θεὸς at 3:16) and gendered language (γυναῖκας at 3:11)output/translations/mark_07.json — controversy + healing (36 verses), good for OT-citation tracking (Isa 29:13 LXX-dependence at 7:6-7) and Aramaic transliteration policy (Κορβᾶν, ἐφφαθά)Read its check report at output/check_reports/<book>_<NN>_review.md — the audit trail for that chapter (key-term consistency, TNBT structural diff, OT citation acknowledgment, synoptic parallels, back-translation similarity).
That's the minimum reviewer package. You can stop there or go deeper.
New Testament — COMPLETE. All 27 NT books shipped (~7,950 verses across 260 chapters), tagged at the book-<slug>-v1 level following per-book end-of-book audits. Matthew · Mark · Luke · John · Acts · Romans · 1–2 Corinthians · Galatians · Ephesians · Philippians · Colossians · 1–2 Thessalonians · 1–2 Timothy · Titus · Philemon · Hebrews · James · 1–2 Peter · 1–3 John · Jude · Revelation.
Old Testament — complete (39 books, 929 chapters). Every book from Genesis to Malachi is source-complete and has been through the end-of-book editorial audit + external-AI cross-review (39/39). The whole Bible was aligned to BSB (English) versification in July 2026; scripts/check_eremos_bundle.py is the ship gate that enforces that structure.
Corpus total: the complete canon — 1,189 chapters, translation-complete and live in the Eremos app (31,086 verses, BSB-numbered).
📖 Read it: every shipped book has a flowing-prose reader edition at output/reader/<book>.md. Examples — matthew.md, mark.md, luke.md, john.md, acts.md, romans.md, revelation.md, ruth.md, jonah.md, genesis.md. The same translation also surfaces verse-by-verse with tap-to-reveal scholarly notes in the Eremos reader app (web at eremosapp.com; iOS via TestFlight; Android via Play Console).
All shipped chapters pass the relevant ship-gate cadence (NT-side: key-term consistency, TNBT structural, OT citation acknowledgment + DB-drift detector, synoptic parallels, back-translation, thai_summary coverage, claim-consistency / hallucination detector, Greek-field integrity, multi-word phrase consistency. OT-side: Hebrew-field integrity, divine-names lock, MT-anchored versification, Rachasap honorifics-binding, back-translation, summary coverage, and a claim-consistency / hallucination detector wired advisory-first). Currently 18 check scripts under scripts/check_*.py; the orchestrator (run_checks.py) routes by language. Optimal-equivalence polish happens in the translation session itself (there is no API pipeline in this project) and proposes idiomatic refinements without auto-applying.
Reader-facing Thai context summaries (thai_summary field) appear on roughly half of all verses — well above the 30–50% target in docs/THAI_SUMMARY_STYLE.md. NT→OT citation database at data/nt_ot_citations.json curates ~720 entries as of NT-v1. Every commit signed; every translation file SHA-256 fingerprinted in HASHES.md. End-of-book audits land an editorial review packet + external AI cross-review packet (Grok / ChatGPT / Gemini) before book-level v1 tags ship.
Our Greek source is the SBL Greek New Testament (SBLGNT, ed. Michael W. Holmes, 2010), licensed CC BY 4.0. At Mark 1:1, SBLGNT's main text reads Ἀρχὴ τοῦ εὐαγγελίου Ἰησοῦ ⸀χριστοῦ. — the ⸀ marker flags υἱοῦ θεοῦ ("Son of God") as a variant in its apparatus. The phrase is present in Codex Vaticanus (B), Bezae (D), Washingtonianus (W), and most Byzantine manuscripts; absent from Codex Sinaiticus's original hand (א*), Theta, and citations by Origen. NA28 / UBS5 bracket it with a {C} certainty rating. Homoioteleuton (accidental drop from five consecutive -ου genitives) is the most commonly proposed mechanism for early omission.
The BSB — which we use only as an English reference bridge, not as a source — uses a different eclectic text that includes the phrase without brackets.
Our approach: translate from SBLGNT as stated, and document every divergence from other major traditions in the verse's notes. Our Mark 1:1 note records the manuscript evidence explicitly. This same editorial principle handles every contested passage — Mark 1:41 (ὀργισθείς vs σπλαγχνισθείς), Mark 9:29 (with-or-without "and fasting"), Mark 14:24 (with-or-without "new"), and the ⟦double-bracketed⟧ longer ending of Mark 16:9-20.
| Book | Status | Notable features |
|---|---|---|
| Mark | ✅ NT-v1 | Pilot book — corpus methodology emerged here. SBLGNT-omitted "Son of God" at 1:1 bracketed; ὀργισθείς/σπλαγχνισθείς at 1:41; ⟦double-bracketed⟧ longer ending 16:9–20 |
| Matthew | ✅ NT-v1 | Royal-register (ราชาศัพท์) consistency; Sermon on the Mount; Olivet Discourse; Passion + Resurrection. Late retag book-matthew-v2 applied ἄφεσις-cluster phrase-lock |
| Luke | ✅ NT-v1 | Magnificat / Benedictus / Nunc Dimittis; parable-dense chs 10–19; Emmaus road; end-of-book external Gemini + Claude cross-AI review pass; Western non-interpolation table for Luke 24 |
| John | ✅ NT-v1 | I AM (ἐγώ εἰμι) discourse vocabulary lock; pericope adulterae 7:53–8:11 in ⟦double brackets⟧; high-Christological prologue treated as one register-unit |
| Acts | ✅ NT-v1 | Pentecost speech with composite OT citations (Joel 2 / Ps 16 / Ps 110); Stephen's Heilsgeschichte rehearsal; Pauline missionary speeches calibrated to Athenian/Roman audiences |
| Romans | ✅ NT-v1 | δικαιοσύνη / δίκαιος family lock per dikaioo_dikaiosyne_family_2026-05.md; Romans 16:25–27 doxology disposition documented |
| 1 Timothy | ✅ NT-v1 | Inaugural pilot of the optimal-equivalence rules; ὃς/Θεὸς textual variant at 3:16; γυναῖκας crux at 3:11 |
| Hebrews | ✅ NT-v1 | ἱλαστήριον disambiguation Heb-vs-Rom per hilasterion_heb-rom_distinction_2026-05.md; LXX-citation pattern preserved verbatim where the Greek text quotes from LXX |
| Revelation | ✅ NT-v1 | Heavenly liturgy register; Trinitarian title-cluster locks; ἔρχομαι ταχύ register-split per erchomai_tachy_revelation_register_split_2026-05.md |
| Ruth | ✅ OT pilot | First OT book — established חֶסֶד → ความรักมั่นคง corpus-lock; Boaz's gate-scene legal vocabulary; redeemer-kinsman thread |
| Jonah | ✅ OT pilot | Tetragrammaton convention validated in narrative + psalm + prophetic-oracle voices; MT/English versification map for ch 2 (great-fish boundary); Exod 34:6–7 attribute-formula recitation lock; sailor + Ninevite parallel-conversion arc |
| Genesis | ✅ book-v1 | Full primordial history + patriarchal narrative (50 ch); OT-narrative register and divine-name compound handling (יהוה־אלהים in the Eden narrative) established here |
| Job | ✅ shipped | Wisdom poetry (42 ch); El/Eloah/Shaddai in the dialogue vs the Tetragrammaton in the prose frame; God-as-adversary lament register (16:9, 30:21) preserved in Rachasap, not softened; chaos-monster transliterations (Rahab/Leviathan/Behemoth); MT↔English versification seams at chs 40–41; goel → พระผู้ไถ่ at 19:25 |
| Daniel | ✅ shipped | Mixed Hebrew + Aramaic sections; foreign-emperor royal register vs Hebrew-vizier plain register; apocalyptic visions; Son-of-Man at 7:13 |
Each translation lives at output/translations/<book-slug>_<NN>.json with full Greek/Hebrew source, BSB English reference, Thai rendering (thai), optional literal alternative (thai_literal), reader-facing Thai context summary (thai_summary), key decisions with rationale, and notes on textual variants / hapax legomena / OT citations / Tetragrammaton handling. OT verses additionally carry a versification sub-object in known-divergence zones.
Source materials (all open-licensed — see ATTRIBUTIONS.md):
Translation engine: Claude (Anthropic) — latest Opus with 1M context window and max effort/thinking, fresh chat per chapter. Back-translations are hand-written by Claude in the translation session and saved to output/back_translations/ — no script calls the Anthropic API and no API key exists in this project. See RULES.md §11 for current model assignments.
Pipeline (see docs/TRANSLATION_WORKFLOW.md):
scripts/next_chapter.py reads data/production_order.json) OR specify the book/chapter explicitlyscripts/extract_book.py auto-routes OT codes to extract_book_hebrew.py)scripts/enrich_with_uw.py) — produces book-intro + per-chapter notesRULES.md, the relevant translator-decision docs, and the most-recent prior chapter output for style continuitykey_decisions rationaleoutput/back_translations/scripts/run_checks.py) — currently 18 distinct check scripts under scripts/check_*.py; the orchestrator picks the relevant subset for NT vs OT chapters. Iterate via scripts/revise_chapter.py if any check fails (max 3 passes before human review)scripts/ship_chapter.sh BOOK CHAPTER — gates on green checks, regenerates HASHES.md + reader doc + plain-text + feedback markdown, commits to main, and (at end-of-book) auto-launches the editorial-review subagent + opens the audit PRscripts/ship_book.sh BOOK — rebuilds the Eremos app bundle, opens + auto-merges the EremosVercel2 PR, bumps iOS CURRENT_PROJECT_VERSION, uploads to TestFlight via altool, and tags book-<slug>-v1. Android Play Console upload is a separate manual step.Eremos Translation is an evangelical Protestant Bible translation project. To be specific:
If you're a Thai pastor reviewing the translation: this is meant to sit comfortably alongside the modern Thai Bibles you already trust — same canon, same family of source text, same translation philosophy as the evangelical Bibles on your shelf. The differentiators are licensing (CC0) and process transparency (every decision audited in git), not theology.
Thailand already has excellent modern Thai Bibles, but they are copyrighted. Eremos Translation is different in three specific ways:
output/translations/*.json.It is not a competitor to the established Thai Bibles; it's an alternative for use cases that require open license (apps, education, derivative works, redistribution). The translation philosophy is "optimal equivalence" — same as BSB in English.
If you're qualified to give technical feedback, these are the dimensions that matter:
glossary.json for the corpus-wide ledger.scripts/check_*.py) the right ones? What checks are missing?output/translations/*.json. Does the documented rationale make sense? Could you trace the decision back to the Greek?| File | Purpose |
|---|---|
RULES.md | Canonical translation rules, sources, philosophy, check mandate |
GENRES.md | Genre-aware production plan + recommended translation order |
ATTRIBUTIONS.md | Source materials and their licenses (CC BY 4.0, CC-BY-SA, CC0) |
LICENSE | CC0 1.0 Universal Public Domain Dedication for our output |
NOTICE | One-line summary of CC0 + dependency licenses |
glossary.json | Auto-maintained per-lemma rendering ledger across all translations |
data/production_order.json | Canonical 1,189-chapter translation order (NT complete + OT in progress) |
data/nt_ot_citations.json | NT→OT citation database (curated, ~720 entries) |
data/synoptic_parallels.json | Synoptic parallel-passage map for cross-gospel consistency |
data/versification_map.json | OT MT-vs-English verse-numbering divergence map (Psalms superscriptions, Joel 3, Mal 3/4, Jonah 2, etc.) |
data/hebrew_idioms.json | Running ledger of Hebrew idioms encountered + their Thai handling |
docs/TRANSLATION_WORKFLOW.md | Step-by-step process for adding a chapter |
docs/translator_decisions/ | Per-decision lock docs (chesed, divine-names, register, leitwort handling, etc.) |
docs/end_of_book/ | Per-book end-of-book audit packets (editorial review + external AI handoff) |
output/translations/<slug>_<NN>.json | Per-chapter Thai translations with rationale |
output/back_translations/<slug>_<NN>.json | Thai → English back-translations for the check |
output/uw_notes/<slug>_FRONT.md | unfoldingWord book-level scholarly intro |
output/uw_notes/<slug>_<NN>.md | unfoldingWord per-chapter scholarly notes |
output/check_reports/ | Audit trail for each chapter's check cadence |
scripts/ | Extract, enrich, translate, check, ship pipeline |
Public (since 2026-04-17, when the Gospel of Mark reached first-draft completion). The CC0 license applies to all output in output/translations/ and the curated databases under data/. Source material under sources/ is git-ignored — each source has its own license and must be cloned separately from upstream (see ATTRIBUTIONS.md).
NT-v1 complete; OT well underway (20 of 39 books). All 27 NT books carry per-book book-<slug>-v1 tags, each landed after a green ship-cadence + an end-of-book editorial-review subagent + at least one external AI cross-review pass. Human-review remains ongoing through the in-house review form (see §10b of RULES.md); verse-level revisions can land at any time and trigger a book-<slug>-v2-style retag.
On the OT side the project has moved well past the original Ruth + Jonah + Genesis bootstrap: all 39 OT books are source-complete on main (929 chapters). Each book ships chapter-by-chapter, then earns its book-<slug>-v1 tag after its end-of-book audit + external-AI review. The whole canon is translated; current work is reviewer-driven polish and cross-book consistency dispositions.
Human-review intake: translation reviewers can sign up at eremosapp.com/review — a bilingual (Thai + English) form backed by the project's question-bank (per-file YAMLs at EremosVercel2/shared/review-questions/). Reviewers are screened by the project lead and matched to question categories per their declared skill set (Thai naturalness / theology / biblical-language exegesis / reader comprehension / app UX).
Companion app: Eremos — surfaces the translation in a Thai Bible reader with a "Translation Notes" popup showing rationale + a one-tap "Flag error" feedback button. Web at eremosapp.com; iOS via TestFlight; Android via Play Console.
Project lead: Ben VanScyoc (benvanscyoc@gmail.com) — missionary in Thailand, DMIN student.
Feedback, corrections, theological review, and Thai-native naturalness review all welcomed. Open an issue or email directly.
CC0 1.0 Universal (Public Domain Dedication)
The Eremos Translation — the Thai translation text and associated rationale, decisions, and notes produced by this project — is dedicated to the public domain under the Creative Commons CC0 1.0 Universal Public Domain Dedication.
To the extent possible under law, the contributors to this project have waived all copyright and related or neighboring rights to the Eremos Translation output (the content of output/translations/*.json files and the Thai text in the distributed bundle).
This work is published from: Thailand (and anywhere else contributors may reside).
You can copy, modify, distribute and perform the work, even for commercial purposes, all without asking permission.
Full legal text: https://creativecommons.org/publicdomain/zero/1.0/legalcode
IMPORTANT SCOPE LIMITATION:
The CC0 dedication applies ONLY to the Eremos Translation's own output:
It does NOT apply to any source materials we reference or depend on. Those retain their own licenses. See ATTRIBUTIONS.md for the full list of source materials and their respective licenses.
This listing does not have a supported local package template. Use the maintainer’s documentation for its hosted endpoint, authentication, and client-specific setup. No install command has been inferred.