Data Sources & Citation Policy

What we use: The Sefaria Open API powers all live text fetching in the Talmudy Reader. Every daf (page) of Talmud Bavli is fetched in real time and cached via our backend proxy. We also use Sefaria's linked commentaries API to surface related Rashi, Tosafot, Maharsha, and other traditional commentators to the Ask AI engine.

Datasets: Sefaria's full dataset (texts, links, names, topics) is available on HuggingFace (sefaria org) and via their GitHub export. Talmudy uses the live API rather than a static dump to ensure texts stay current with Sefaria's corrections and additions.

License: CC BY-NC 4.0 for all open-licensed texts. Commercial use requires a separate agreement with Sefaria. Talmudy operates under educational, non-commercial use in compliance with this license.

What we use: The English translation displayed in the Talmudy Reader is the Koren Noé Talmud, translated by Rabbi Adin Even-Israel Steinsaltz z"l. This is the most comprehensive, beginner-accessible English Talmud translation available in digital form.

Aramaic text: The original Aramaic with full nikud (vowel diacritics) is also from the William Davidson digital edition, with vocalization data contributed by Dicta — Israel Center for Text Analysis.

License: CC BY-NC 4.0 via Sefaria. © Koren Publishers Jerusalem. Original Babylonian Talmud (Aramaic/Hebrew) redacted c. 500 CE — public domain.

What we use: Dicta contributed the nikud (vowel points / diacritics) to the William Davidson vocalized Aramaic text. These vowel marks are what make the KANDLES phonetic analysis possible — we strip the nikud to isolate root consonants and assign phonetic family colors. Dicta's work is foundational to the KANDLES system.

What it is: KANDLES is Talmudy's proprietary phonetic color-encoding system. Each of the 7 phonetic consonant families in Biblical Hebrew/Aramaic is assigned a color:

KANDLES grid data is stored in the Talmudy repository and is not derived from any third-party phonetic dataset. The encoding maps words positionally — grid tile N corresponds to word N in the Aramaic text of that amud.

What we use: The Ask AI — Study Partner feature on every reader page is powered by Google Gemini 1.5 Flash. Questions are sent to Gemini together with the Sefaria text of the current daf and a list of relevant traditional commentaries (Rashi, Tosafot, Maharsha, and others fetched from the Sefaria links API).

Data pipeline: User question → Sefaria text fetch → linked commentary summary → Gemini prompt → structured response with sources cited. No user questions are stored or used for model training.

Important disclaimer: Ask AI is an educational study aid only. It is not a halachic authority. For practical rulings or religious questions, consult a qualified rabbi.

Sefaria publishes their full text corpus — all 3,000+ Jewish texts including the complete Talmud Bavli and Yerushalmi, Mishnah, Midrash, responsa, and more — as open datasets on HuggingFace. These are the same datasets that power the Sefaria API. Talmudy currently uses the live API rather than a local dataset copy. Future AI fine-tuning for Talmud-specific question answering will be explored using these HuggingFace exports.