वेदान्वय

VedAnvaya — The Vedas, connected.

Four Samhitas held as one corpus. Read any verse in Sanskrit with its recitation and its translation, follow a deity or a formula across the collections, and see the evidence behind every connection.

Read the Vedas Open the graph

20,210 canonical mantras · 4 Samhita corpora · 17,780 recitations

Open the graph

What this atlas holds.

Read from the knowledge service when this page was built, and shown with the clause each one needs to be read correctly.

Canonical mantras: 20,210Every verse addressed by a canonical key and a canonical citation.
Samhita corpora: 4Rigveda, Samaveda, Yajurveda, Atharvaveda — one recension each.
English translations: 18,427A corpus’s own published English, plus 194 renderings reused from a parallel verse and labelled as reuse.
Verse recitations: 17,780One recording per verse, mapped to an exact verse and streamed from its source.
Samavedic notation witnesses: 1,136Verses carrying source-explicit svara marks, validated against the printed witness.
Cross-Veda links: 8,723Measured across six corpus pairs, of which 6,582 are literal textual reuse.
Named subjects in the graph: 157deities
616seers
4,825formulas
575metres
103rites
Counted apart rather than summed: one subject can be a metal, a substance and a concept at once, so a single total would count it three times.

Corpus, translation, subject and cross-Veda figures are read live from the knowledge service. The notation figure comes from the certified completeness record, from the copy pinned in this build because the service did not answer. How each figure is established.

श्रुतिThe Vedas

Four Samhitas, one recension each.

Each collection keeps its own structure, its own recension and its own words for both. Open one to read it.

The four collections held, with the recension of each, its verse count, and how many verses carry a translation and a recitation.
Collection	Recension	Verses	Translated	Recited	Read
Rigvedaऋग्वेद	Śākala	10,552	10,480	10,552	Open Rigveda
Samavedaसामवेद	Kauthuma (ārcika)	1,844	reused renderings only	none released	Open Samaveda
Yajurvedaयजुर्वेद	Śukla, Vājasaneyi Mādhyandina	1,975	1,950	1,777	Open Yajurveda
Atharvavedaअथर्ववेद	Śaunaka	5,839	5,715	5,451	Open Atharvaveda

Each edition’s recension, and what it excludes, is set out in full on the scope page.

RV 1.1.1

Script: Romanised, accented
Witness: not read
Translation: not read

A verse, and everything standing behind it.

The Sanskrit, the recitation, the translation, the seer and the deity the traditional index assigns, the words the verse actually contains, and the wording it shares with another collection.

Each of those comes from a named layer, and each says how it was established. An assignment made by the index is never shown as a statement the Sanskrit makes.

Open this passage

17,780 verses, each with its own recitation.

One recording per verse, catalogued from its publisher and streamed from there rather than copied. A recording is attached only where our canonical key lands in the source's numbering and the Sanskrit matches this corpus's text, because playing the wrong verse is worse than playing nothing. That is a checked mapping and not an audible review; how far each has been taken is on the scope page.

Rigveda10,552 of 10,552
SamavedaNo catalogued recitation
Yajurveda1,777 of 1,975
Atharvaveda5,451 of 5,839

अन्वयConnections

The same wording stands in more than one collection.

6,582 cross-Veda links are built in typed kinds rather than as one similarity score: exact parallel, near parallel, variant, directed reuse, shared formula, shared entity vocabulary. Which kind a link is decides what it can be used to argue.

720 formula families are traced through the corpus, and each one carries the collections it reaches. Every link opens onto both verses, so the resemblance can be read rather than taken on trust.

Follow the wording, including 107 families that reach all four

A Rigvedic verseRV 1.1.1, the text as GRETIL prints it
stands again in the SamavedaMatched on wording, classified, and kept distinct from mere reuse
and the difference is shownBoth texts side by side, with what each witness actually reads

Ask a question. Check the answer.

Retrieval runs first, over a fixed catalogue of channels, and the model sees only what retrieval returned.

Every sentence carries its evidence

Each factual claim comes with a citation you can open: the retrieved item, its Sanskrit, its translation and its canonical citation. The evidence is the answer’s working, not a reading list attached afterwards.

And a question it cannot reach is refused

On its 60-question benchmark the current build returned 60/60 acceptable outcomes, nothing misleading and nothing invented, with 17 questions honestly refused rather than answered from a gap.

Ask a question

मन्त्रRV 10.129.7

i̱yaṁ visṛ̍ṣṭi̱r yata̍ āba̱bhūva̱ yadi̍ vā da̱dhe yadi̍ vā̱ na | yo a̱syādhya̍kṣaḥ para̱me vyo̍ma̱n so a̱ṅga ve̍da̱ yadi̍ vā̱ na veda̍ ||

He, the first origin of this creation, whether he formed it all or did not form it, Whose eye controls this world in highest heaven, he verily knows it, or perhaps he knows not.

Translated by Ralph T. H. Griffith, 1896Open this passage

Every claim says how it was established.

Four grades, attached to the claim rather than assumed from context. A weaker grade is never promoted by sitting next to a stronger one.

Source: The text, an edition, or a statement the traditional index makes.
Derived: Computed from a source by a stated rule, the same way every time.
Model-assisted: Extracted by a language model, kept as a candidate, always marked.
Interpretation: A reading of the data, recorded with what would falsify it.

Where a layer was never built, the answer says so instead of returning a zero that a reader could mistake for silence in the Vedas.

Read the evidence model See the scope of this edition

Where to start.

Built by one person over the sources named on the sources page, across 18,427 translated verses. About this project