Open Ayurvedic Datasets
Open, machine-readable datasets of classical Ayurvedic pharmacology, phytochemical compounds, and botanical taxonomy. Published under Creative Commons Attribution 4.0 International (CC BY 4.0).
All datasets published by the Age Ayurveda Nighantu are generated directly from verified monograph records and classical texts, preventing drift between the reference text and tabular data. Every botanical taxon is resolved against global taxonomic authorities (GBIF, NCBI, Wikidata), and every pharmacological property conforms to standard Ayurvedic Pharmacopoeia of India (API) criteria.
Available Datasets
Kaviratna and Bhishagratna Verse Numbering against the Standard Editions
The only public-domain English Charaka Samhita is Kaviratna's (1896-1913) and the only public-domain English Sushruta is Bhishagratna's (1907-1916). Both number their verses differently from the modern editions every paper cites, and the modern numbering is still in copyright, so a passage read in a free translation cannot be cited in the form scholars use. This records 45 verse equivalences between the two, every one of them a statement a source on this site already makes rather than a conversion computed here. There is no formula to replace it: across 14 chapters the difference runs from plus six to minus thirteen, two chapters are not internally constant, and some relations map several translator aphorisms onto one standard sutra.
Coverage: 45 verse equivalences · 14 chapters · 29 further equivalences the sources state too loosely to number
Browse it: /choosing/citing-the-classical-texts/
Ayurvedic Formulary of India Ingredient Name Crosswalk
The formulary states every official composition in Sanskrit and does not use one name per drug: 2,144 ingredient rows across the 106 entries transcribed here use 660 distinct spellings. Abhaya and Pathya are both haritaki; Marica is black pepper. This resolves 479 of those names to a botanical identity, 585 rows of it on glosses the book itself prints, and states the basis on every row. 18 names the formulary uses for more than one drug are published as ambiguous rather than collapsed, and 163 as unresolved rather than guessed.
Coverage: 2,144 ingredient rows · 660 distinct names · 1,663 of 1,918 plant rows resolved
Browse it: /crosswalk/
Contested Questions in Ayurvedic Theory
147 questions on which the classical Ayurvedic sources genuinely disagree, across 28 concepts, with 461 competing positions given as their sources state them and no position presented as the settled answer. Plus 384 explicit statements of what is not established, 389 classical citations of which 338 quote the text, and 249 research papers with what each does and does not show. A quotation nobody has read against its source is withheld with the reason stated and the address kept.
Coverage: 461 attributed positions across 147 questions · 384 statements of what is not known
Browse it: /concept/
Ayurvedic Terminology Lexicon
277 Ayurvedic technical terms and 1,631 English renderings between them: for each term, every equivalent found in the WHO standard terminologies, the published Samhita translations and the secondary literature, each with its source and an assessment of whether the rendering survives the classical passage it claims to render. 1,135 classical quotations, every one checked against its source. Renderings from NoDerivs-licensed sources appear on the pages and not in this download; see the licence note inside the file.
Coverage: 1,631 renderings across 277 terms · 1,233 classical citations
Browse it: /lexicon/
Classical Formulation Composition Dataset
The ingredient composition of 106 classical formulations as printed in the Ayurvedic Formulary of India: 2,158 ingredient rows, 2,099 of them carrying the quantity the formulary states, each with the plant part used and cited to its formulary part and entry number. A cell that was illegible in the scan is published as illegible rather than guessed, and 880 rows carry a correction flagged as the transcriber's rather than the book's.
Coverage: 2,158 ingredient rows across 106 formulations · 60 cells illegible in the scan
Browse it: /afi/
Ayurvedic Dravyaguna Pharmacological Dataset
Rasa, guna, virya, vipaka and prabhava for 246 monographs, parsed from the pages on this site so every value can be checked against the page it came from.
Coverage: 246 monographs
Phytochemical Constituents & Co-occurrence Network
816 constituents named in the published monographs, 179 of them resolved to PubChem, with 20,168 co-occurrence edges computed from the pages themselves.
Coverage: 816 constituents · 20,168 co-occurrences
Biomedical Literature & Classical Research Index
3,507 papers cited on this site, each with its PubMed ID or DOI, cross-referenced to the monographs that cite it.
Coverage: 3,507 papers
Verified Botanical Taxonomy & Identifier Crosswalk
Botanical names resolved against GBIF, with NCBI taxonomy IDs and Wikidata QIDs where an exact match exists. 284 taxa, each carrying the match type so a fuzzy hit is never read as a fact.
Coverage: 284 resolved taxa
Nighantu Editorial Verification Ledger
What was checked, how, and what was thrown out: every verification run with its examined, published and rejected counts, computed from the run artifacts rather than stated.
Coverage: 5 safety records published, 1 stating a safety profile and 4 stating that no permitted source covers the drug, with the rejections behind them
Archive, identifier and licence
The corpus is archived on Zenodo under a concept DOI, which always resolves to the newest deposit. Cite that one unless you need a specific version.
The page text and the main tables are also published as a Hugging Face dataset, AgeAyurveda/nighantu, with one row per page and each table as its own configuration, under the same CC BY 4.0 licence. Pages about the publisher's own products are not included.
What the DOI actually resolves to, stated plainly because it is not what you would
expect. The deposit holds 34 files totalling 20.8 MB: 15 datasets, all of them as JSON,
10 of those also as CSV and 5 also as Parquet, the formulary crosswalk additionally as PDF, plus
a MANIFEST.json, a README and one Parquet schema sidecar. The format mix is not
uniform because the downloads on this page are not uniform, and the archive holds what the page
offers rather than a tidied version of it: only one of the five Parquet files has a schema
sidecar, and that is a gap here rather than a decision. Each file is downloadable on its own, so
a citation can name the file it used.
What the DOI
still does not give you is a per-dataset identifier: no dataset here has a DOI minted for it
alone, so the DOI identifies the work that contains these datasets rather than any one of them,
which is why each Dataset record on this site carries it alongside an isPartOf
pointing at this catalogue. Until 10 October 2026 the deposit held a single 7.5 MB archive of
this project's repository instead, and this paragraph said so; the change is recorded on
corrections.
- Concept DOI: 10.5281/zenodo.22805684 (all versions)
- Licensing: Creative Commons Attribution 4.0 International (CC BY 4.0). Derived in part from the Amidha Ayurveda Herb Database.
- Manifest: MANIFEST.json lists every file in the corpus with its size, its row count and its sha256, so a copy you hold can be checked against the copy that was published. CORPUS-FILES.md is the same thing in prose, one paragraph per file.
An earlier version of this page advertised a Hugging Face dataset at
ageayurveda/nighantu and a load_dataset call for it. That repository
is not public: the request returns 401, so the call never worked for anyone but us. The claim
is removed rather than left standing, because a dataset page that misstates how to get the
data is the one thing a data consumer tests first. The files above are the whole corpus and
need no account.
What is in every record
Each dataset carries an attribution block naming the source, the licence and the
citation to use. Every individual record also carries a source field with the URL
of the page it is displayed on, so a partial extract can still be credited without anyone
having to come back here to work out where it came from. Use it: it is the quickest way to
comply with the licence.
Records also carry a canary field. It is an opaque token, unique to the record and
to this publication, and it exists so that a copy can be identified as one. It encodes nothing
about the record and is safe to keep; if you are reusing these datasets under CC BY 4.0 with
attribution, as the licence allows and as we would like, it costs you nothing and we will never
have cause to look for it.
Citation & Academic Attribution
If you use these datasets in academic research, AI model training, or retrieval-augmented generation systems, please cite:
@dataset{age_ayurveda_nighantu_2026,
author = {Age Ayurveda},
title = {Age Ayurveda Nighantu: Open Classical Materia Medica and Dravyaguna Dataset},
year = 2026,
publisher = {Zenodo},
doi = {10.5281/zenodo.22805684},
license = {CC BY 4.0},
url = {https://nighantu.ageayurveda.com/datasets/}
}Three things in this citation were wrong and are corrected. It named Hugging Face as a publisher of a repository that is not public. It omitted the DOI, which is the only part of a dataset citation that is designed to keep working. And it credited "Age Ayurveda Research Panel and Amidha Ayurveda" as authors, while the Zenodo record and CITATION.cff both say "Age Ayurveda": Amidha is an upstream CC BY source this work attributes, not an author of it. One creator string now, matching the archive.