Search ancient texts, then read them in layers.

What it is and how it works

The README calls it a search-first library and reader for ancient texts. It is a Next.js (App Router) and TypeScript app with Tailwind, PostgreSQL through Drizzle, and Better Auth for accounts. The planned screens are home search, library catalog, search results, a work landing page and the layered reader.

1. Import with provenance in the repo

   Importer scripts fetch each source and store works, witnesses (one version of a text, such as a root or a translation), divisions and passages. Every witness records its source, an SPDX licence, attribution text and the import time. The docs say sources with unknown rights block ingestion.

   Source: docs/CORPUS.md, src/db/schema.ts, src/scripts

2. Keep layers aligned in the repo

   Each passage has a layer (original, transliteration or translation) and a language. For SuttaCentral, the root and the translation are fetched together under the same segment IDs, so the layers stay aligned. Passage text is stored raw so it stays citable, with a separate normalized search form.

   Source: docs/CORPUS.md, src/db/schema.ts (passages)

3. Search, then read in the repo

   Search runs on PostgreSQL with diacritic-insensitive matching. Results open the reader for a witness at /read/[witnessId], with stacked or parallel layers and commentary, parallels and footnote panels.

   Source: README.md, src/components/PassageLayers.tsx, src/components/reader

4. Ask, with checks needs a model

   Ask is a retrieval-augmented chat. It embeds the question, recalls passages by vector and keyword search, runs a bounded tool loop, then checks quotes against the text and citations against what was retrieved before showing a trust badge. It needs an embedding and chat model, for example Ollama or another OpenAI-compatible API.

   Source: docs/ASK.md

Features

• Library and works

  Catalog, work pages, authors, sources and collections.

  src/app/library, works, authors, sources

• Layered reader

  Original, transliteration and translation layers, a witness selector and division navigation.

  src/components/ReaderView.tsx

• Parallels

  Panels that show related passages in other works.

  src/components/reader/ParallelsPanel.tsx

• Ask

  Grounded Q&A with citations that link to passages and a stable sutraya: URN. Modes scope answers to a tradition.

  docs/ASK.md

• Translate and OCR

  A translation helper with an AI drafting dialog, and an OCR page that sends PDFs or images to an external OCR service.

  README.md, src/app/translate, src/app/ocr

• Saved, history, workspace

  Saved items, reading history, documents and sheets, each scoped to a user.

  src/app/saved, history, workspace

• MCP server

  Tools to search the corpus, read a passage, get parallels and resolve a URN, over the read API.

  mcp-server/

• Local or hosted

  Single implicit local reader by default, or multi-user accounts with usage limits for AI features.

  README.md "Auth"

• Retrieval checks

  A regression test set and gate for Ask.

  eval/, npm run eval:gate

The layered reader, as a mock

No screenshots exist in the repository. This is an illustrative mock, not a screenshot. It follows the passage model: one row per passage, with an original layer and a translation layer sharing a position, and notes beside them. The text is a short sample, the notes are placeholders.

Illustrative mock, not a screenshot.

Corpus and licences

The first corpus is a curated set of Pali root texts, and Bhikkhu Sujato's segmented English translations, from SuttaCentral. The repository's corpus notes record both as CC0-1.0. Other sources the importers can load carry different licences, so check the source before reusing any text.

Source: docs/CORPUS.md. This page repeats the repo's records and has not re-checked them with the publishers. Corpus text is credited to SuttaCentral and its translators.

Status

• in the repo code and docs describe it
• needs a model works only with a model or service you supply
• planned a roadmap or gap list item
• unverified not run for this page

• Local prototype in the repo

  Setup is Node 20 or newer, Docker for Postgres with pgvector, then npm run setup and npm run dev, on port 3010. The app was not started for this page.

• Ask needs a model

  Requires an embedding model and a chat model that you provide. Answer quality was not tested here.

• OCR needs a model

  Calls an external OCR service, with an optional vision-model fallback for images.

• Hosted prototype unverified

  The docs describe a hosted prototype and a multi-user mode. Whether one is live now is not stated, and no public address is given here.

• Wider corpus planned

  Gap lists for East Asian and Saiva sources, Zen texts and rights decisions are in the docs. Several items are marked as pending a licence decision.

• Tests unverified

  Unit tests, a Playwright smoke test and a retrieval eval exist. None were run for this page.

Links

The README does not give a public address for the project, so none is claimed here.

• Source repository

  Local folder sutraya. Start with README.md, then docs/CORPUS.md, docs/ASK.md and docs/ROADMAP.md.

• Corpus source

  SuttaCentral, credited above. Its site is named in the repository's corpus notes.