Skip to content

Repository navigation

Open Granth v1

Latest
Latest

Choose a tag to compare

View all tags
@satnam satnam released this 17 Aug 04:00
· 5 commits to main since this release
v2026.08.16
This tag was signed with the committer’s verified signature.
satnam Satnam Narang
GPG key ID: 4BEAA6C5AB3F24C4
Verified
Learn about vigilant mode.
9dc39b3
This commit was signed with the committer’s verified signature.
satnam Satnam Narang
GPG key ID: 4BEAA6C5AB3F24C4
Verified
Learn about vigilant mode.

Sri Guru Granth Sahib Ji in AI-ready Markdown.

General-purpose AI systems can produce fabricated Gurbani. Sometimes a real verse is given the wrong ang; sometimes whole passages are invented and presented as scripture. The MCP server gives an AI client a way to retrieve Gurbani from the released source files with ang and line citations instead of asking a model to reproduce it from memory. Claimed quotes can be checked against those same files.

Browse Sri Guru Granth Sahib Ji at opengranth.org.

Contents

  • Corpus: all 1,430 angs (60,555 primary text entries), one Markdown file per ang, with Gurmukhi, romanized transliteration, English, and per-file provenance frontmatter.
  • Static site: every ang browsable, search across Gurmukhi, transliteration, and English, plus a verification page. The core reading, search, and verification functions use only same-origin released files and do not depend on third-party content or APIs.
  • MCP server: seven tools (lookup_ang, search_gurbani, get_verse, search_by_raag, search_by_author, random_ang, verify_gurbani), running locally against the bundled corpus over stdio.
  • Verification: verify_gurbani checks claimed quotes against the released source files. Accepted verification inputs receive verified, partial_match, or not_found; matches include source locations. English and transliteration use normalized whole-line and containment matching; fuzzy fallback is limited to Gurmukhi.

Getting started

The corpus is plain Markdown; clone and read. For the MCP server and the test suite, see the README. The server runs locally over stdio and finds the corpus bundled beside it, so no configuration is needed.

Provenance

Corpus derived from @shabados/database 5.0.0-next.0, selecting its SGGS source. Within that upstream database, the primary Gurmukhi rows used by the import come from the SSA2 asset, identified there as Shabadaarth and attributed there to the SGPC Committee, Sri Amritsar. The DSSK asset supplies the English layer, attributed there to Dr. Sant S. Khalsa. Open Granth generates the Roman layer locally.

The Roman line is mechanically generated as an approximate companion to the Gurmukhi, not an authoritative pronunciation guide. Within Open Granth, the Gurmukhi is the source layer; the Roman and English layers are reading aids. See NOTICE and docs/TRANSLITERATION.md for complete attribution and the transliteration scheme. The exact upstream assets, versions, and row counts are recorded in docs/provenance.md.

Verification record

  • Release gate: 49/49 passed, run from this tree.
  • Test suite: 84 passed (includes phrase-boundary regression tests for Gurmukhi search and exact-line verifier regressions).
  • MCP acceptance: all seven tools exercised over stdio in a clean environment against the bundled corpus.

Versioning

Open Granth uses calendar versioning: a release is identified by the date its snapshot was finalized, YYYY.MM.DD, tagged vYYYY.MM.DD. A date names the exact snapshot of the corpus, site, verifier, and MCP tooling that a verification claim was checked against. This release names the snapshot finalized on August 16, 2026.

Dedication

Development of Open Granth began independently in January 2026. The initial release candidate was completed on August 12, 2026.

In memory of my father.

ੴ ਸਤਿ ਨਾਮੁ

Assets 2
Loading