The best place to learn to read Hebrew, biblical and modern, through comprehensible input.
You learn a language by reading things you can nearly understand, in quantity. targum.page makes Hebrew readable at every level: a chapter of Ruth, this morning's news, a novel you brought yourself, each shown sentence by sentence beside a translation, so you always understand what you are reading.
Tap any word and targum gives you its dictionary form, so the word you meet in Genesis is the same word when it turns up in a sports report. Say how well you know it, and the page quietly clears as you learn: what you have yet to meet is lit, what you are working on turns leaf, what you know reads plain. Verbs come with their conjugations. Names and numbers are marked as what they are.
You can also ask. targum has a chat that answers from the library and from your own
shelf: what to read next at the words you know, what you have read, what a text you are
looking at will cost you in words you have not met. It does not spend on your behalf —
when something would have to be built, it says so and stops. The same tools are in Claude
and ChatGPT if you would rather ask from there: add targum as a connector and it answers
about your own shelf, your own words and the library, wherever you already ask. targum mcp does the same over stdio for a client on your own machine.
What sets it apart is that everything is curated to what a learner needs. A library that runs from Tanakh to this week's news, sorted by how much of it you can already read, with a published translation wherever one exists and AI translation, sentence by sentence, in context, for everything else. Your own texts welcome: an EPUB, an article, a link, a recording. One vocabulary across biblical and modern Hebrew, held by dictionary form, so every word you learn in Tanakh is yours in a newspaper too. Vowel points where the text has them, guessed elsewhere, with every guess marked. Progress that counts real things, words known, texts finished, days read, and places you on a level ladder: the ulpan for Hebrew, the CEFR for French, Russian and Italian. Readers that are one file each: a phone, an e-reader, offline, in pages.
The hosted version at targum.page is in alpha and invite only. Once invited, sign in with an email address and a link.
uv tool install "targum[difficulty]"
export ANTHROPIC_API_KEY=...
targum serve
Drop in an EPUB, a text file, an audio file, or a link — a podcast episode counts. The library of texts with published translations that targum.page carries is data, kept privately alongside the code.
AGPL-3.0-or-later. Copyright © 2026 David Langellotti.
Some of the models targum ran were licensed for non-commercial use only: the forced
aligner behind the speech-align extra, and Stanza's models through the treebanks they
are trained on — Hebrew's, and until 2026-09-13 English's and Russian's as well. None has
run since. The AGPL cannot lift somebody
else's NonCommercial term, so LICENSING.md sets out what was encumbered and what
replaced it.
Contributions are welcome. CONTRIBUTING.md explains the sign-off and what it grants.