- Starting price
- Free
- Free tier
- Yes
- Platforms
- Web browser - Mobile web - Offline-capable
- Developer
- SIL Global (SIL International)
- Launched
- -
- Updated
- Aug 5, 2026
The verdict
Scripture Forge is the best free tool of its kind, and it is not for you unless you are on a Bible translation team with a Paratext registration - the login is Paratext, the project list is your Paratext project list, and only a Paratext admin can connect a project. If you lead a team with at least one Gospel finished, it is close to a default choice. If you want to read, study or memorise Scripture, this is the wrong page.
Try Scripture Forge ↗Opens scriptureforge.org
Scripture Forge is a free web application from SIL Global for teams doing Bible translation. It does three things: it lets a team draft and edit together in a browser, it runs community checking sessions where ordinary speakers of the language answer comprehension questions about a passage, and it generates rough machine drafts of books the team has not reached yet, trained on the books it has already finished. All of it synchronises two ways with Paratext, the desktop translation environment the industry runs on.
The listing that sent you here calls it a free, open-source web app for collaborative Bible translation with AI-powered draft generation. That is accurate, and the open-source part checks out - SIL's terms state that most of its software is offered under an MIT open source license, with source at github.com/sillsdev. What the listing leaves out is the entry requirement. SIL's help pages describe exactly one way in: click 'Log in with Paratext' or 'Sign up with Paratext', then supply the email address tied to your Paratext Registry account. The My Projects page lists the Paratext projects you already belong to, and connecting a new one is restricted: 'Only admins can connect a project on this page.' There is no path here for a curious individual to start a translation project of their own.
That framing matters, because the interesting question about Scripture Forge in 2026 is not whether it is good. It is what happens when a machine translation system produces first drafts of Scripture in languages with very few speakers, across hundreds of projects at once. SIL is unusually direct about this, and the honest thing is to quote it rather than characterise it. Understanding Draft Generation states: 'All of the drafts that are created contain errors that need to be corrected by the translation team. For some projects, the quality will be too low to be useful.' The New Testament guidance adds that drafts 'are not perfect and will always need review and improvement by experienced translators.'
So the product is narrow, the audience is small, and the vendor is candid. What follows covers what the platform does mechanically, what powers the drafting, what SIL commits to on data handling, and where its documentation contradicts itself - which it does, on the most practical question a team lead asks.
✓ The good
- Free with no tier structure - SIL publishes no pricing page for Scripture Forge; /pricing, /plans, /subscribe and /give all return 404, and the terms cap SIL's total liability at 'THE AMOUNT YOU PAID SIL TO USE OUR SERVICES', which for this product is nothing.
- Two-way Paratext synchronisation - community checking answers push back into Paratext as notes under their own note tag, so consultants can filter them in or out of the notes pane they already work in.
- Community checking is built for the field - mobile-first, with audio recordings of both questions and answers, and playback of the passage with each verse highlighted as it is spoken, so speakers who cannot yet read the language can take part.
- Answer-blinding is the default - when a checker is presented with a question, existing answers from other checkers are hidden so they will not influence the answer given; they appear only after the checker responds.
- Offline work is supported - SIL states team members can work offline, with changes saved once a connection is re-established, which matters in the places these projects run.
- Draft quality is estimated before you rely on it - book and chapter estimates use a green, yellow, red scheme, where green means the draft 'is likely to be useful' and red means it 'may not be useful'. SIL offers most of its software under an MIT open source license, and Serval, the drafting API behind this one, is a public SIL project rather than a vendor black box.
✗ Watch out
- A Paratext registration is effectively mandatory - every documented login path runs through Paratext, and only a Paratext project admin can connect a project. There is no standalone Scripture Forge project.
- The documentation contradicts itself on how much translated text you need - the published answers range from an entire New Testament to a single completed book.
- The best drafting workflow is not self-service - SIL states plainly that 'Incremental Draft Generation is not yet available as a self-service Scripture Forge feature, and must be requested through the Scripture Forge EITL team', EITL being SIL's Engineer-in-the-Loop support staff.
- Drafting access is granted by application - teams click 'Sign up for Drafting' and fill out a form; only back translation projects are enabled by default.
- The privacy policy is generic rather than product-specific - it names Language Depot five times and Webonary twice, mentions Scripture Forge only in the page furniture, and never describes project confidentiality or where project text is stored.
- Uploading grants SIL a broad content licence - 'a non-exclusive worldwide license to use, host, store, reproduce, modify, create derivative works, communicate, publish, publicly perform, publicly display and distribute such content'.
Best for
- Paratext project administrators
- Translation team leads and consultants
- Community checking coordinators
- Field teams on low-bandwidth connections
Avoid if
- Personal Bible reading or study
- No Paratext registration
- Sermon preparation and commentary work
- Language learning or vocabulary drilling
What Scripture Forge is
SIL describes Scripture Forge as 'an online app for team-based Bible translation, featuring AI help and tools for community review.' It sits alongside Paratext rather than replacing it. Paratext is the desktop translation environment used across the industry, developed and funded in partnership by United Bible Societies, SIL Global, Biblica, Wycliffe Bible Translators, ETEN and Mission Mutual. Scripture Forge takes a registered Paratext project, mirrors it in a browser, and adds three things Paratext does not do well: browser editing for members without the desktop software, structured community checking, and machine-generated drafts.
The audience is therefore institutional. A team already inside the Paratext ecosystem gains a web surface, a checking workflow and a drafting engine at no cost. Everyone else stops at the login screen, by design rather than oversight.
The drafting engine, in SIL's own words
Draft generation runs through Serval, SIL's open source REST API for natural language processing. SIL's Serval page states it currently supports 'Neural machine translation (NMT) for back translations into more than 200 languages based on the NLLB-200 model from Meta', alongside statistical machine translation, and lists MADLAD-400 and large language models under Future Development rather than current capability. Serval's documented dependencies are MongoDB, ClearML for GPU training jobs and S3 for training data and models. So: a Meta open-weights translation model on SIL infrastructure, not a general-purpose chatbot writing Scripture.
What it is trained on is the team's own work. SIL explains it as a two-step process - learn the language, then translate the text - where 'the system learns by seeing the same sentence written in multiple languages, one in a language that it already understands, and one in the language you are translating into.' The parallel text is the team's reference translation on one side and their completed books on the other. It is not learning theology; it is learning a mapping into one target language from that team's own finished chapters.
Community checking, mechanically
A project administrator enables community checking when connecting the Paratext project or later from Settings. Questions are created one at a time or imported in bulk, each scoped to a verse or verse range by typing a start and end reference. A question can carry a recorded or uploaded audio version of itself, and the project can hold audio recordings of the passage with timing files so each verse is highlighted as it is spoken. That is what makes the feature usable with speakers who cannot yet read the language.
Checkers arrive two ways: an email invitation carrying the Community Checker role and an invitation language, or a share link that admits anyone who clicks it. Answers land in a Questions and answers view where each is marked for export or resolved, with a filter for unreviewed answers. Exported answers sync into Paratext as notes carrying the original question, the checker's name and the answer, under a distinct note tag consultants can filter or hide. One documented gap: audio answers appear in Paratext but cannot be played there, and must be opened back in Scripture Forge.
How much translated text you actually need
This is where SIL's own pages disagree, and it is the question a team lead needs settled before planning a year of work. The Understanding Draft Generation help page says: 'At this time, we recommend that you have at least the New Testament translated.' The SIL AI project page sets a numeric threshold: 'When a project contains about 8000 verses across multiple genres - equivalent to a complete New Testament - draft generation can provide a draft of the next book.' The later New Testament guidance says something materially different: 'At least one completed Gospel is required to get started', and its FAQ goes further - 'New Testament draft generation is only available after you have translated at least one book in your project.'
Report both figures, do not average them. Matthew is 1,071 verses against the roughly 8,000 SIL cites for a full New Testament - about an eighth. The most defensible reading is that 8,000 verses describes when drafts become reliably good and one Gospel describes when the feature will run at all. SIL supplies the calibration: with only the Synoptics done, 'drafts of John tend to be of a lower quality than drafts of the Synoptics, and drafts of Acts are usually the lowest quality of these books.'
Sequencing and the Engineer-in-the-Loop
SIL publishes three book sequences - Canonical, Luke/Acts/5 T's, and a Draft-Optimized order designed to smooth out quality swings. The Draft-Optimized epistle order is specific to the point of being startling: 1JN, 2JN, 3JN, 1TH, 2TH, REV, 1CO, 2TI, GAL, 1PE, PHP, ROM, TIT, EPH, PHM, JAS, 2PE, JUD, 1TI, COL, HEB, 2CO. Hebrews and 2 Corinthians are held back because they draft worst, and Acts is recommended before any Epistle because it affects epistle quality more than any Gospel does.
Incremental draft generation - draft a whole book, hand-improve 8 to 10 chapters, then regenerate the rest from the improved text - is the technique that most improves output on the Gospels and Acts. It is also the one you cannot run yourself: SIL states it 'is not yet available as a self-service Scripture Forge feature, and must be requested through the Scripture Forge EITL team.' The same Engineer-in-the-Loop team supplies chapter-level quality estimates and flags chapters likely to come out badly. Drafting is a supported service with a human queue attached, not a self-serve button.
Pricing
Free
Free
Everything described here - collaborative editing, two-way Paratext sync, community checking with audio, and AI draft generation - is available at no charge, with no paid tier published anywhere on scriptureforge.org. SIL Global funds it as part of its language technology work, and Paratext, which Scripture Forge depends on, is developed and funded in partnership with United Bible Societies, SIL Global, Biblica, Wycliffe Bible Translators, ETEN and Mission Mutual. The real cost is not money but eligibility: a Paratext Registry account, admin rights to connect a project, and an approved drafting application.
There is no price. SIL publishes no pricing page, and the obvious URLs - /pricing, /plans, /subscribe and /give - all return 404 rather than redirecting to a donation ask. The terms state that most SIL software 'is offered under an MIT open source license', with source code at github.com/sillsdev.
Funding is institutional rather than transactional. SIL Global develops Scripture Forge inside its language technology programme; Paratext, which the platform depends on, is developed and funded in partnership with United Bible Societies, SIL Global, Biblica, Wycliffe Bible Translators, ETEN and Mission Mutual. Serval, the API behind drafting, is a separate SIL AI project. None of them bills the translation team.
The cost that does exist is administrative. You need a Paratext Registry account to log in, admin rights to connect a project, and an approved drafting application - teams click 'Sign up for Drafting' and complete a form, with only back translation projects enabled by default.
Read the liability clause before assuming free means risk-free. SIL provides the service 'AS IS', excludes all warranties to the extent permitted by law, may 'suspend or stop any given software or service altogether', and caps its total liability at the amount you paid - zero. It does commit to reasonable advance notice and instructions for preserving your data if a service is discontinued.
Where Scripture Forge falls behind
No product-specific privacy policy. The policy linked from Scripture Forge's footer is SIL's general software privacy policy; its body never mentions Scripture Forge, discussing Language Depot and Webonary instead. It confirms that 'some of our services are hosted by Amazon' and limits sharing outside SIL to consent, administrators, contracted service providers and legal reasons. It does not tell a team in a restricted region where their draft text sits, how long it is retained, or whether a project can be marked private. The nearest answer is in the help FAQ: 'Scripture Forge maintains a version of the project on SIL-controlled servers, and is responsible for keeping that project secure', with the caveat that 'the data does travel over the Internet when you connect to the Scripture Forge interface online.'
Access control leans on link sharing. Confidentiality follows Paratext project membership, which is the right model, but community checkers are invited by email or by a share link that admits anyone who clicks it. The only documented control is a single switch - 'Allow community checkers to invite community checkers' - and turning it off means 'only project admins will be able to add community checkers to the project'. There is no documented middle setting letting translators invite checkers without enabling every checker to do the same. In a sensitive region, a circulated link is the weak point, and you get a switch rather than granular control.
The content licence is broader than most teams will expect. Uploading data grants SIL a non-exclusive worldwide licence to use, host, reproduce, modify, create derivative works, publish, publicly perform, publicly display and distribute that content, for the limited purpose of improving SIL's language support and services. Structurally that is hosting boilerplate, but it is not narrow, it applies to unpublished Scripture drafts in minority languages, and it runs 'as long as you have content hosted on the service.'
The best drafting workflow still needs a human queue. Incremental draft generation and chapter-level quality estimates both route through SIL's Engineer-in-the-Loop team rather than the interface, and drafting access itself is granted by application. Generation takes roughly an hour and a half and 'can sometimes take much longer depending on server availability and the number of books requested', by SIL's own account. This is a shared research cluster and behaves like one.
No launch year and thin published evidence. SIL nowhere states when Scripture Forge launched, and the only effectiveness figure it publishes is one comparison over September 2024 to September 2025: projects using draft generation drafted and reviewed an average of 1,000 more verses than projects that did not. One year, one metric, no error bars, and no published measure of how much of that text survived checking.
Scripture Forge vs. Paratext vs. Parabible
These three are not competitors so much as three jobs that get confused with each other. Paratext is the professional translation environment: the project database, the checks, the consultant workflow, the registration that certifies a translator's involvement in an approved project. Scripture Forge is a free browser layer on a registered Paratext project, adding community checking and machine drafting. You do not choose between them - Scripture Forge does not function without a Paratext registration behind it.
Parabible is the tool most readers of this site actually want when they land on a page like this: a free browser interface to the Hebrew and Greek text with morphological tagging, no institution, registration or project required. If your interest in Bible translation is curiosity rather than professional responsibility, Parabible is the correct destination and Scripture Forge is not.
The line to draw is responsibility. Scripture Forge assumes you are accountable to a language community for a text it will read for decades, that you have consultants and checking cycles, and that a machine draft is a starting point you are qualified to correct. SIL's framing matches: it calls the platform 'a tool designed to help you navigate the drafting phase more effectively and quickly while maintaining the quality of your work', and adds that 'it is not a replacement for the translation steps.'
The bottom line
Scripture Forge is the strongest free tool in its category, and almost nobody reading this should open it. If you administer a Paratext project with at least one Gospel completed, it costs nothing, syncs cleanly, gives you a real community checking workflow with audio, and can save meaningful drafting time - SIL's figures put projects using draft generation about 1,000 verses ahead over a year. Without a Paratext registration there is no version of this you can use. And whichever side of that line you sit on, the sentence worth carrying away is SIL's rather than ours: every draft the system produces contains errors a translation team has to correct, and for some projects the quality will be too low to be useful at all.
Alternatives to Scripture Forge
ParabibleFree browser access to the Hebrew and Greek text with morphological tagging. No account, no institution, no project - the right tool for someone who wants to read the original languages rather than produce a translation.
STEP BibleTyndale House's free study platform with original-language tools, lexicons and interlinear display. Aimed at study rather than translation production, but far more useful to a general reader than a translation workbench.
BiblingoPaid biblical-language learning app built around vocabulary and grammar acquisition. Relevant if the reason you landed here is wanting to learn Greek or Hebrew rather than to run a translation project.
Bolls BibleFree, fast multi-translation reading site with a public API. Useful for anyone who wanted a lightweight way to compare translations and landed on a Bible translation platform by mistake.Frequently asked questions
Is Scripture Forge really free?
Yes. There is no pricing page on scriptureforge.org and the usual pricing URLs return 404. The terms state that most SIL software is offered under an MIT open source license, and they cap SIL's total liability at the amount you paid to use the service, which here is nothing. SIL Global funds the platform as part of its language technology work.
Do I need a Paratext licence to use Scripture Forge?
In practice, yes. Every login path SIL documents runs through Paratext: click 'Log in with Paratext', then supply the email tied to your Paratext Registry account. My Projects lists the Paratext projects you belong to, and 'Only admins can connect a project on this page.' Community checkers invited by link or email join a project someone else already connected. There is no documented way to create a Scripture Forge project that does not originate in Paratext.
What AI model generates the drafts, and what is it trained on?
Drafting runs through Serval, SIL's open source NLP API. SIL's Serval page states it supports neural machine translation into more than 200 languages 'based on the NLLB-200 model from Meta', alongside statistical machine translation, with MADLAD-400 and large language models listed as future development rather than current capability. Serval's documented dependencies include ClearML for GPU training jobs and S3 for training data and models. The training data is the project's own parallel text - the team's reference translation matched against the books that team has already completed.
How accurate are the AI drafts?
SIL publishes no accuracy percentage, and its own language is deliberately cautious. Understanding Draft Generation states: 'All of the drafts that are created contain errors that need to be corrected by the translation team. For some projects, the quality will be too low to be useful.' The New Testament guidance says drafts 'are not perfect and will always need review and improvement by experienced translators', and names a concrete failure mode: a project moving from the Gospels to Acts may produce poor translations for new words such as Pentecost and Cappadocia. SIL's stated measure of success is usefulness to the team, 'not their ability to stand alone as a finished product'.
How does community checking actually work?
A project admin enables it in settings, then adds questions scoped to a verse or verse range, optionally with recorded audio. Checkers join by email invitation or share link and answer in a mobile browser view; the passage can play as audio with each verse highlighted as it is read. Other checkers' answers stay hidden until a checker responds, unless the admin changes that. Answers land in a review queue to be marked for export or resolved, and exported ones sync into Paratext as notes under their own note tag.
Where is project data stored, and can a project be kept private?
SIL's help FAQ says Scripture Forge 'maintains a version of the project on SIL-controlled servers', that the project also syncs to Paratext servers, and warns that 'the data does travel over the Internet when you connect to the Scripture Forge interface online.' The linked privacy policy is SIL's general software policy: it confirms some services are hosted by Amazon and limits sharing outside SIL to consent, administrators, contracted service providers and legal reasons, but never addresses project confidentiality. Access follows Paratext project membership plus invitations; a share link admits anyone holding it, and admins can disable checker-initiated invitations.
How much translated Scripture do I need before drafting is worth it?
SIL's pages give different answers and both are worth knowing. The help centre recommends 'at least the New Testament translated', and the SIL AI project page puts a number on it - about 8,000 verses across multiple genres. The newer New Testament guidance says 'At least one completed Gospel is required to get started', with its FAQ stating drafting is available after one translated book. Matthew alone is 1,071 verses against roughly 8,000 for a full New Testament, so the gap is real. Treat one Gospel as the point where the feature runs, and 8,000 verses as the point where output becomes dependably useful.
Sources & further reading
- Scripture Forge Help centre - login, Paratext connection, community checking, FAQs ↗
- Scripture Forge Help - Understanding Draft Generation ↗
- Scripture Forge Help - Guidelines for New Testament Draft Generation ↗
- SIL AI and NLP - Scripture Forge project page ↗
- SIL AI and NLP - Serval project page ↗
- Scripture Forge Terms of Use ↗
- Scripture Forge Privacy Policy ↗
- Paratext - About, development and funding partners ↗
More Greek & Hebrew Tools
Blue Letter BibleThe serious-study site for everyone who can't afford Logos - and a surprising number of people who can.
Bible HubThe most-cited free Bible study site on the internet - a thirty-translation parallel view, a Strong's-tagged interlinear, and a wall of classical commentaries on every verse, all one click apart and all completely free.STEPBibleA free, open-source original-language Bible study site built by actual textual scholars at Tyndale House Cambridge - and the best-kept secret in serious Bible study.
BibleArcThe diagramming workbench seminarians swear by and most casual readers have never heard of - and the reason that gap exists is the same reason its fans will not give it up.Scripture4AllA quiet, donor-run project that gives the whole Bible away as a word-for-word interlinear - Hebrew and Greek under a literal English gloss, with grammatical parsing on every word, free and downloadable.BiblingoBiblingo teaches biblical Hebrew, Greek and Latin by immersion rather than by parsing charts, and it charges for it. Fifty-two lessons per language, a Hebrew University doctorate behind the curriculum, and one uncomfortable fact - every tool it competes with is free.