Both a study and publishing tool

Build your own dictionary — and keep every word you study.

Lexifye is where you build your own dictionary: the words you keep meeting in the text, what you found in them, and why it mattered. Look one up today and it is still there in five years — with everything you have added since. When it is ready to leave your desk, the same entries become a printed book, an EPUB, or a private reader you share with a study group or a whole congregation.

There is a real pleasure in this that nobody warns you about. A dictionary you wrote yourself is not a chore you maintain — it is the most enjoyable thing on your desk, because every entry is a small victory you get to keep. Ten words become thirty; the thirty start talking to each other; and one morning you realise you are reading with something no study Bible could have handed you, because you built it.

Sign in once from Claude, DeepSeek or your own agent, and it writes entries into your dictionary on your behalf — under your name, and revocable from the same page.

Building with an AI? The tool surface is the manual, written to be read by a machine: 87 tools, each with its access level and full schema, plus the authoring conventions and the markup grammar. You do not need any of it to use the web editor.

This is what comes out

Not a mock-up — the actual PDF Lexifye typeset from 168 entries, the same way it will typeset yours. Hebrew and Greek in their own faces, transliteration beside the lemma, every word that has an entry of its own quietly linked to it, and a full index at the back.

The same entries also become an EPUB, a reader that works on a phone, and a 6×9 interior ready for a printer. You write once.

Download the sample PDF

Covenant Words — A Lexifye Showcase
Jørn Andre Halseth · 168 entries · 62 pages · A4 · rendered by this service

From the blog

All posts →

Why build or join in on a dictionary at all

Because the looking-up is not the study. Anyone can find that hesed is "steadfast love" — and find it again next year, and the year after, having kept nothing. The study is what you noticed the fourth time you met the word, in a passage nobody told you to connect it to.

A lexicon of your own is where that accumulates. Every entry holds your senses, in your words, with the references that convinced you. The Hebrew and Greek you cite become real entries themselves — native script, transliteration, a composed gloss drawn from our scripture engine — so a word you half-knew in March is a word you can teach from in October.

You do not need Greek or Hebrew to start, and you do not need to be a programmer to use it: it is a web editor you sign into, and the original-language work is done for you when you cite a word.

A study tool and a press, in one package

Most study software stops at the note. The note is where the work usually dies — readable by you, on one machine, in a format nobody else can open. Lexifye treats the entry as publishable content from the first keystroke: structured senses, typed spans, real scripts. So the step from "my notes" to "a book with a spine" is a render, not a rewrite.

The same dictionary comes out in seven shapes, from one press, every time you ask:

A dictionary here stands entirely on its own — you never have to write a book to get the good of it. But if you do, it becomes something more.

One of three — and complete on its own

Three services built to work together, none of which needs the other two. The joins are settings, not dependencies: disconnect any of them and what you made stays yours, stays readable, and keeps rendering.

THE SCRIPTURE ENGINE Darash Hebrew and Greek, morphology, Strong’s, the classical lexicons you cite a word back as an entry YOUR DICTIONARY Lexifye chesed ḥesed your senses, your sources, every version you have had — and the press for all of it a work is built only what it uses THE WRITING Junifye a book, a study, next week’s handout — its glossary written THE SCRIPTURE ENGINE Darash Hebrew and Greek, morphology, Strong’s, the classical lexicons you cite a word back as an entry YOUR DICTIONARY Lexifye chesed ḥesed your senses, your sources, every version you have had — and the press a work is built only what it uses THE WRITING Junifye a book, a study, next week’s handout — with its glossary already written
Both arrows are one-way pulls. Lexifye never writes into Darash, and a Junifye work never edits your dictionary.

Darash — where the languages come from

Cite H2617 in a definition and Lexifye asks Darash what that is, then writes the answer into your dictionary as a real entry: the pointed lemma in Hebrew script, an SBL transliteration, and a gloss composed from the lexicons — Strong's, Brown-Driver-Briggs, Abbott-Smith, Liddell-Scott-Jones, Thayer. You choose how far it goes: the words you actually cited, one related hop, or deeper on request. Nothing it writes can overrule you — a hand-authored entry always wins a collision.

Turn it off (depth 0) and your dictionary is unchanged; you simply write the original-language part yourself.

Lexifye — where the words live

This service, and the middle of the three by design: it is the system of record for lexical content in the whole ecosystem. A definition is authored here and nowhere else, and everything downstream holds a cache rather than a copy — so when a word changes here, what quotes it is stale by definition and is told so.

It never needs either neighbour: a private lexicon of one reader's study is a finished thing, and it prints.

Junifye — where it ends up in print

Point a work at a dictionary and its glossary builds itself on every render: the terms the text uses, plus the terms your own definitions lean on, all the way down until nothing dangles. It records the exact version of each definition it took, down to a checksum — so you can prove what a word meant in the edition someone is holding — and an entry keeps the anchor it was published with, whatever you rename afterwards.

Point it at nothing and it is simply a place to write.

Not one of the three, but next door: build your own Bible with Tringine — choose a translation, the scope and a trim size, and get a print-ready PDF and an EPUB3. All four are made by Publifye AS, a Norwegian publishing house in Oslo (org.nr 826 774 622, founded 2016), which publishes Bibles, builds the tools they are studied with, and runs infrastructure for churches.

One word, all the way through

What actually happens, in the order it happens, with the real text at each step. Nothing here is aspirational — it is the path chesed took through the sample you can download at the top of this page.

  1. You write the definition, and cite the word

    In the editor, in ordinary prose, with one tagged span for the Hebrew:

    [h:H2617:ḥesed:חֶסֶד] denotes the *loyal, active love* that
    binds parties in covenant — kindness that far exceeds
    what mere obligation could require.

    That is the whole of the markup grammar you need for this. The round trip is exact: what you read back is what every later read gives you.

  2. Lexifye asks Darash what H2617 is

    And writes the answer into your dictionary as a real entry — the pointed lemma חֶסֶד, the SBL transliteration ḥesed, and a gloss composed from the lexicons. It is marked as machine-made, it is capped, and it can never overwrite a word you wrote yourself.

    You set how far this goes, from off to deep. Stop citing a word and its auto entry is reconciled away; your own entries are never touched.

  3. The dictionary sets itself

    Every occurrence of covenant in that definition becomes a link to the covenant entry, because this dictionary has one. The headword prints with a capital and stays stored as you typed it. It files under C in the letter index, and the index at the back lists every entry, each one a destination and a way back.

    A screen PDF in light or dark, a 6×9″ interior with running guide words, an EPUB3, a web reader, JSON — or the raw LaTeX, if you want to take it elsewhere.

  4. A Junifye work takes only what it needs

    Write a study of Ruth that uses the word, and its glossary appendix appears with chesed in it — along with covenant, because your definition of chesed leans on it, and so on down until nothing dangles. The build records the checksum of the exact version it used.

    Edit the dictionary afterwards and the book already in print is unaffected: an entry keeps the anchor it was published with.

What the glossary actually does

That last step is the one people ask about, so here it is in full. Connect a Lexifye dictionary to a work in Junifye and the glossary writes itself. Not "export your terms and paste them in" — the work pulls the entries it actually needs, every time it is built. That is true of a full book and equally true of a six-page document: a study of one book of the Bible, a topical study on covenant or atonement or the fear of the Lord, a handout, a paper, a series outline. Same dictionary behind all of them.

So the fourth thing you write is faster than the first, not slower — because the words are already yours, already defined, already typeset.

Built a word at a time, over months

This is not a project you sit down and finish. It is a thing you add to on a Tuesday because a word stopped you, and again three weeks later when the same word turns up somewhere you did not expect — and the second entry is better than the first because you had both passages in front of you.

That is the whole point, and it is why it has to be yours. A study of Romans that took you a winter, a topical study on covenant you have been circling for a year, the words a single Gospel keeps returning to — none of that fits in a session. It accumulates. Six months in you are not looking things up any more; you are reading with your own work open beside the text.

And you decide, later and separately, who else ever sees it: keep it private and it is a lexicon of one reader's study; share it with a group — your class, your elders, a field team — and everyone works from the same senses; publish it as a book, an EPUB or a reader and it goes to anyone you want. Nothing forces that decision at the start, and nothing stops you changing it.

It does not have to be only you

A dictionary is the rare kind of work that gets better when a second person is in it — and worse when that person has to be sent a file and asked to write comments in a margin. So there are two different ways to let somebody in, and knowing which one you want is most of it.

One person, one dictionary

A guest editor is added to a single dictionary and nothing else. Your supervisor, a colleague on the same corpus, a fellow student a chapter ahead of you, the brother who actually reads Hebrew. They get the entries, the history and the reader in the tool where the work lives — so when they find the error they fix it instead of describing it. Take the access back and it is gone on their next click.

A standing group

Your elders, a seminar, a research team, a translation committee, a board. Attach the group and every member is an editor of that dictionary; add somebody to the group and they have it. Switch on autoshare and the group picks up each new dictionary you start, so you stop granting access one dictionary at a time.

Or an audience, who just read

A class, a congregation, a cohort of pupils. They need nothing from you but the file: the EPUB, the reader that works on a phone, or a printed copy each. Nothing to sign into, nothing to install, no account, and no limit on how many.

Nothing anybody does is unrecoverable. Every change is stamped with the name of the person who made it in that dictionary's own log, every definition keeps twenty versions with a diff and a revert, and a deleted entry sits in the trash for thirty days. You can hand somebody real access without handing them the power to ruin a year of work — which is the thing that actually stops people collaborating.

There is no charge for any of it. Your plan counts dictionaries, not people.

Work with an AI in it, not beside it

Most people using an assistant on a piece of research are doing it in a second window, pasting things back and forth and losing half of it on the way. Lexifye is built the other way round: the dictionary an agent can reach is the same one you are looking at.

Point one at it and it can search your entries, look a lemma up, draft a definition in the same markup you write by hand, and fix a term across four hundred definitions in one pass — showing you exactly what it would change before it changes anything. It signs in as you, so it can do what you can do and nothing more: your dictionaries, your access, no second account.

And you keep the diff. Every edit an agent makes is versioned and revertable exactly like your own and appears in the log beside everything else. You are not trusting it — you are reviewing it, afterwards, at your own pace.

Private until you decide otherwise

There are no public dictionaries. Nothing here is browsable, indexable or fetchable without a credential — not the reader, not the JSON, not the PDFs. Knowing an id or a UUID is not access. What you write stays yours until you hand it to someone.

When you do, there are two different moves. Bring somebody IN — a guest editor on one dictionary, or a standing group of elders, a class committee or a field team — and they work on it with you, with every change carrying their name. Or hand an audience a COPY: the EPUB, the reader or the printed interior, which needs nothing from them and has no limit on numbers. Nothing is really deleted either: soft deletes with a thirty-day grace window, a restore for every entity, and twenty versions of each definition kept with diff and revert.

Three kinds of caller reach content: authors signed in through Publifye SSO; customers holding an OAuth token from our authorization server; and authorised Publifye services calling service-to-service, which may read and never write. This page, the tool surface, /llms.txt and /health are the only things served anonymously.

What you find out later

Find it, then fix it everywhere

Search a dictionary by any part of a headword or by what a definition actually says — you do not have to remember the exact spelling of a word you wrote in March. And when you change your mind about a rendering, one literal find-and-replace runs across every definition at once, with a preview before anything is written. It never touches a Strong's code, a reference or a link target, only prose. Every change it makes is revertible.

How it is set is yours

A4, Letter or 6×9″. The body size the typesetter can actually deliver — it refuses a size it would have to fake. Entries running on, or a page each for a dictionary of long articles. Linking on or off, and either every occurrence of a word or just the first mention in an entry. These belong to the dictionary, not to us: no setting here is a decision a renderer makes behind your back.

Notes that never leave your desk

Private notes on the dictionary or on a single entry — the reading you rejected, the passage you still need to check, the question you want to put to someone. Visible to editors, never rendered into any format, never exported, and they follow an entry into the trash and back out again. Working notes, not content.

Under the hood

Lexifye is also the Publifye ecosystem's system of record for lexical content: a definition is authored here and nowhere else, and other services hold caches — never copies of the truth. When a word changes here, everything downstream is stale by definition, and is told so.

Dict → Entry → Definition

A dictionary holds lemmas; a lemma holds one or more senses. A sense is structured content — blocks of typed spans — not a blob of prose, which is what lets the same definition typeset correctly as a PDF, an EPUB and a web page.

87 atomic MCP tools

Every mutation is one small, validated, individually-authorised call. Errors are written for the agent that has to recover from them: each names the field, the fix and a minimal valid call.

Round-trippable markup

Definitions have a compact text form with the guarantee parse(render(x)) == x — the round trip is exact on the parsed document, not on your bytes. One normalising pass may tidy the text itself (\##, [H00002617][H2617]); after that it is a fixpoint, so what you read back is what every later read gives you.

Publication-grade rendering

Screen PDF in light and dark, a 6×9″ print interior with running guide words, EPUB3, an HTML reader, JSON and the raw LaTeX — across 16 scripts, behind a glyph gate that fails a render loudly rather than shipping tofu.

Strong's enrichment

Original-language words cited in a definition become real entries — native lemma, SBL transliteration, composed gloss — pulled from our scripture engine and reconciled as you edit. A hand-written entry always wins a collision.

Locking and history

Every content edit carries proof of the version it was based on, so concurrent editors cannot silently clobber each other. Twenty versions of each definition are retained for seven days, with diff and revert.

Groups and guest editors

Add one person to one dictionary as a guest editor, or attach a standing group so every member is an editor of it. A group can adopt each new dictionary you make automatically, and a dictionary can be handed over outright to somebody already working on it. Membership is resolved at check time, so revoking is instant rather than eventual.

Nothing is really deleted

Deletes are soft by default with a thirty-day grace window and a restore for every entity, and every change is attributed in a per-dictionary audit stream.

Durability in three layers

Redis for the working set, content-addressed blobs and per-dictionary exports on disk, and an add-only object-storage mirror behind that. The restore path is a test that runs on every commit, not a document.

Getting in

Sign in with a Publifye account and start a dictionary — the editor is the whole of what you need, and nothing on the technical pages is required to use it. What each level of access includes, and what it costs, lives in the store: prices belong in one place, and a number copied onto a page like this one goes stale without anybody noticing.

Working with an AI instead of, or as well as, the editor? Connect it takes a couple of steps, and your agent can read the whole contract at the tool surface before you hold a credential.

Calling it

Lexifye speaks MCP over JSON-RPC 2.0 at https://lexifye.publifye.pro/mcpconnect it to your AI in a couple of steps. An agent should read the tool surface first — it publishes every tool with its access level and its full input schema, plus the authoring conventions and the complete markup grammar, so you can evaluate the contract before you hold a credential.

Publifye services obtain a service-to-service read grant through our internal directory instead of registering as a customer.

About News Tool surface llms.txt Sign in

© 2026 Lexifye — the central dictionary service, by Publifye AS.