Ask a question about Jewish texts.
Get an answer that cites its sources.

SeferAI searches Sefaria's library — Tanakh, Talmud, Midrash, Halakhah, Kabbalah and commentary — and writes you a synthesis with every claim linked back to the text it came from. Then it can read it aloud.

Start by reading a real one

These are unedited answers from the public library, written by SeferAI for people who asked. No account needed to read them.

How it works

Three steps. The first one is the only one that needs anything from you.

1

Ask

In plain English or Hebrew. A verse, a sugya, a theme across sources, or something you have been chewing on all week.

2

Read

A structured answer, with inline citations that link straight to the primary text on Sefaria. You can check every claim.

3

Listen

Turn it into a narrated podcast episode you can play on a walk, or subscribe to in any podcast app. This step needs a free account.

What the search is reading

A question is matched against passages, not against page titles. So how many passages there are, and in which language, is most of what determines whether the right source comes back.

Until September 2026
156,185 passages
English only
Every passage, and the matching itself, in English.
A partial library
About one passage in four and a half of the English texts in these categories, and nothing in Hebrew.
Now
3,802,528 passages
English and Hebrew
683,979 English and 3,118,549 Hebrew, across 1,874 English and 6,201 Hebrew texts.
The whole library, in these categories
Every English and Hebrew text Sefaria publishes in the fifteen categories SeferAI searches.
Where the depth wentTexts (EN+HE)Passages
Commentary — Rishonim, Acharonim, modern6,1162,365,241
Halakhah383227,592
Talmud187204,158
Liturgy87196,662
Chasidut162170,009
Responsa163140,279
Jewish Thought, Kabbalah, Midrash, Musar449349,912
Tanakh, Targum, Mishnah, Tosefta, Second Temple528148,675
Why Hebrew matters even if you ask in English. Matching is done by a model that reads the question and the passage in the same space. The old one was trained on English alone, so a Hebrew question was being matched by a model that had never really read Hebrew — and Hebrew-only texts, which is most of the library, were not in the index at all. The new one reads both, so a question in either language can reach a source in either language.

The depth is concentrated where the old index was thinnest: commentary. Asking what the Rishonim make of a verse now has six thousand commentary texts behind it rather than a sample of the English ones.

Two things to know before you start

They are separate choices, and neither one is a trap.

Without an account, your questions are public

You can ask and read without signing in, and it is free. Those questions and their answers join the public study library, where anyone can read them — that library is how most people find this site. Making a podcast is the one thing that needs an account, because it costs real money to narrate.

Signed in, they are yours

Sign in with your email — no password, we send you a link — and your syntheses stay private unless you choose to share one. Signing in is free.

Supporting SeferAI

SeferAI is not trying to make a profit. Every question costs real money to answer — model time, and narration for the podcasts — and support is what keeps it running and free for people who cannot pay.

Free
$0
No account required
  • Ask questions
  • Sourced syntheses
  • Podcast episodes
  • The whole public library
Start asking
Premium
$9.99/month
Cancel any time
  • Premium narration voices
  • Unabridged, full-length episodes
  • Unlimited synthesis
  • Priority processing
Choose Premium

For the curious

How does it find the right sources?

Your question is turned into a vector embedding and matched against a database of Sefaria's texts — about 3.8 million passages, in Hebrew and English, from Tanakh, the Babylonian Talmud, Midrash, Halakhah, Kabbalah and commentary. Matching happens by meaning rather than keywords, so a question about forgiveness surfaces passages that never use the word.

What writes the answer?

A frontier large language model reads the retrieved passages and writes the synthesis from them, citing as it goes. It works from the sources it was given rather than from memory, which is what makes the citations checkable — and why you should check them.

Can I use SeferAI inside Claude?

Yes. SeferAI is available as a connector for Claude, so you can search all 3.8 million indexed passages from inside any Claude conversation and get the references back with links. How to connect it.

What is SeferAI's relationship to Sefaria?

Sefaria is the open library of Jewish texts that SeferAI draws from, and it is a separate organisation. Syntheses are study aids produced by a language model. They are not rulings, and they should be attributed to SeferAI rather than to Sefaria or to any authority quoted inside them.