Source library

Every source this project uses, links to, plans to use, or has decided not to use, with the reason and a direct link. The aim is that anyone, scholar or newcomer, can see exactly where each word comes from and go to the original.

Last reviewed: 1 November 2026 (v2.1.0).

How we choose sources

A source is used only when it meets all five tests:

1. Standing. It is an edition or translation that the tradition or the scholarly community recognises.
2. Permission. Its licence or terms allow this use, and we follow them (attribution, non-commercial, share-alike).
3. Respect. Its authors or publishers have not asked that their work be kept out of AI tools. If they have, we link to them instead.
4. Durability. It is published somewhere stable (a versioned repository, an official mirror), so citations keep working.
5. Verifiability. It has stable identifiers (paragraph numbers, page numbers, URLs), so every passage can be checked.

Status key

Pāli text

English translations

Ānandajoti Bhikkhu: how pages are linked

The index is server/data/anandajoti.json, built by scripts/anandajoti/build.mjs from the site’s “Texts and Translations” section on 1 November 2026 (one page a second, cached outside the repository). Every page is printed line by line, Pāli above English, which lets the build check the text itself:

1. A page is a candidate only for the text the site itself names (its heading or index entry, for example “(MN 146 & MA 146 & AA 1.4.7)”).
2. The page’s Pāli must cover at least 60% of the VRI text of that unit (word 3-grams; the site follows the Sri Lankan Buddha Jayanti edition, so spellings differ slightly), and no other text in the same collection may come within 0.10 of it. Only the Pāli lines (printed in bold) are compared: the English lines and the translator’s notes are set aside first. The VRI’s list-item labels “(Ka)”, “(Kha)” … are dropped, because the site prints lists without them. This catches renumbering and partial translations.
3. In SN and AN the Pāli title printed in the page’s own heading must agree with ours as well (square brackets, the editor’s additions, are ignored). The title is read from the site, never from our own data, and stored with the link as pali_title. A link is labelled number+title+pali only when such a title was read. Never by number alone.

Result: of 166 candidates (80 Udāna suttas, 26 Dhammapada chapters, 19 Pārāyanavagga sections, 10 Bhikkhunīsaṃyutta suttas and 31 other works), 153 are linked and 13 were rejected and left out:

• the site’s “AN 3.62” (its Pāli is AN 3.61 in the standard numbering; its heading title is Titthāyatanasutta, ours for AN 3.62 is Bhayasutta);
• AN 8.52 (the site’s heading title is Bhikkhunovādakasutta, ours is Ovādasutta) and SN 35.95 (the site spells Māluṅkyaputtasutta, ours Mālukyaputtasutta): the Pāli matches, but the SN/AN title rule is kept strict;
• Dhammasaṅgaṇī mātikā, pairs (Ds 1.2, match 0.57): the site follows another edition’s spellings here (for example saññojana where the VRI has saṃyojana);
• DN 22 (the site’s index links only some of its pages, so the build read an incomplete text);
• Ud 4.10, 5.10, 6.6 and 8.10, Dhp 383–423 and Snp 5.3, 5.6 and 5.19, whose match fell below the threshold or was not clearly the best.

An earlier build (first version of this pull request) had a broken pattern for the bold Pāli lines, so it compared the whole page, English included, and took the SN 5 titles from our own data. Both are fixed; the rebuild added Snp 5.2, 5.7 and 5.17 and the mātikā triads and discourse pairs (Ds 1.1, 1.3, once the “(Ka)” labels were dropped) and removed AN 8.52 and SN 35.95. node scripts/anandajoti/audit.mjs re-fetches a fixed sample of 30 links through the server’s own allowlisted client and re-checks the match; on 1 November 2026 all 30 passed. Not linked in this pass: the Mahākhandhaka edition (it covers only the opening sections of Kd 1), the chanting books (excerpts and compilations), the Jātaka pages (they translate the commentary stories, not only the canonical verses), and the Apadāna selection (not attempted).

Retrieval: get_translation with translation_id aj:<page> (for example aj:Udana/1-Bodhivaggo-10) or the translator name. The page is fetched live from ancient-buddhist-texts.net/Texts-and-Translations/ only (the allowlist admits no other path on that host), with the licence and attribution. The translator’s notes are marked [Note: …]. Where a page also translates the commentary (MN 44, MN 98, MN 146, AN 4.70, AN 6.37) those passages are marked [Commentary (Aṭṭhakathā) begins] and the result says they are commentary. Works spread over several pages (DN 16, MN 26, MN 85, MN 118, MN 119, Vibh 6, Vibh 7) are read only as far as the requested characters. A link to any one of their pages (index or part) resolves as a citation of the whole work.

A letter asking Ānandajoti Bhikkhu to confirm the licence and that live retrieval is welcome is drafted, not sent (outreach/anandajoti.md). If he prefers links only, the aj: retrieval is switched off and the links kept.

PTS Creative Commons volumes

The PTS copyright page (checked 1 November 2026) lists, under CC BY-NC 3.0 (10 May 2013): Horner’s Book of the Discipline (6 vols) and Middle Length Sayings (3 vols), Kindred Sayings (5 vols), Gradual Sayings (5 vols), Buddhist Manual of Psychological Ethics, The Book of Analysis, Discourse on Elements, Designation of Human Types, Points of Controversy and Conditional Relations (2 vols); and under CC BY-NC 4.0 (8 April 2021): Ñāṇamoli’s Dispeller of Delusion, The Guide, The Minor Readings and The Illustrator of the Ultimate Meaning, The Path of Discrimination and The Piṭaka Disclosure.

Why none is linked yet:

• No authorised digital text found. The PTS site’s free “Downloads” category lists 24 items (1 November 2026); the 16 on its first page are Pāli texts and studies such as the Jinacarita, Dāṭhāvaṃsa and Pāli Metre, none of them these translations. The copyright page links no files. Scans posted elsewhere by third parties have no stated provenance, so we do not link or read them.
• No stable section addresses. Linking “by section” needs a page or anchor per sutta, rule or paragraph. A scanned PDF gives only volume and page, which we could not check against the text.
• Next step. The PTS letter (drafted, not sent) asks whether there is an authorised digital source. When there is one, the volumes go through the same build-and-audit route as Ānandajoti’s pages: Horner’s Vinaya first, as it would be the first English for the Vinaya here.

Translations in other languages

Surveyed for #35 on 1 November 2026. We link a source only when its licence is published and allows this use, and when its pages can be matched to texts with the same strong evidence as English. Nothing in this table is linked yet. “Not stated” means we found no licence on the pages we checked; it does not mean the work is free.

“Done” for #35 needs at least three languages linked; that waits on licences being confirmed. Letters to these sources are not drafted yet.

Dictionaries and reference works

Commentaries

Parallels, calendars, scripts and practice

Peer projects

We are not the only Buddhist AI project. These are related, and we learn from them:

Tools from other traditions that set the bar we measure against (details in the roadmap): Quran.com MCP, Sefaria MCP, StudyBible MCP, tafsir-mcp.

Mirrors: what happens when a site is down

Why SuttaCentral is linked, not used

SuttaCentral’s data is excellent and largely CC0 or public domain, and it is fully mirrored on GitHub (bilara-data, sc-data). Technically it would have been the most reliable choice. SuttaCentral has, however, asked clearly and repeatedly that its work not be used in AI tools:

• Its licensing page requests that its content “not be scraped or used in any way for the creation of datasets for generative AI or similar”, including by apps built downstream of AI models.
• 41 publication records in bilara-data/_publication-v2.json repeat the request, including Bhikkhu Sujato’s four Nikāyas and Bhikkhu Brahmali’s Vinaya.
• Bhikkhu Sujato has repeated it on SuttaCentral’s forum (2024, 2026).

The request is not legally binding, but a scripture tool for Buddhists should honour the stated wishes of its sources. So we link to SuttaCentral for parallels and other translations and never fetch or bundle its content.

The one remaining use is narrow. The numbering most readers cite (MN 10, SN 46.131–140, pli-tv-bu-vb-pj1) is shared nomenclature, like chapter and verse. To place those numbers on the VRI text, the build once read the file names and opening words of the public-domain Mahāsaṅgīti Pāli. Only the number-to-location mapping and Pāli titles were kept.

What you may do with retrieved text

Full licence notices: THIRD_PARTY.md.

DPPN licence research

Checked on 1 November 2026 for #32. Result: not confirmed as public domain or openly licensed, so no entries are bundled.

What happens now: lookup_name returns no dictionary entry, says why, links to the two online editions and lists the texts of the bundled VRI canon whose Pāli contains the name. Each word of the name is searched separately (Ajātasattu Vedehiputta, Mahā Moggallāna). Texts with the name in the title come first, then texts naming it in the opening paragraphs (the nidāna), then the rest. One-word names of five letters or fewer (Uda, Nanda) are matched as whole words with a case ending; exactWord: true does the same for longer names (Ānanda without ānandajāta, joyful). This is a word search, not a dictionary: the tool says so in its result. The data file is server/data/dppn.json; entries can be added there, each with its citations checked against the bundled canon by tests/dppn.test.mjs, once the Pali Text Society (or Ānandajoti Bhikkhu, for the revised edition) gives written permission. Their reply will be recorded here.

Permission letters

We are writing to each source whose terms need confirming or whose permission we need. The letters, their contact routes and their status are in outreach/. Replies are recorded in this file as they arrive; until then each source is used only as its published terms allow.

Proposing a new source

Open an issue with the source, its licence or terms, where it is published, and what it would add. Check it against the five tests above. New live providers also need bounded retrieval and a live test; see development.md.