The Nisvasa: the oldest tantra, and what it does not yet know

In 1905, in the Durbar Library at Kathmandu, Hara Prasad Sastri catalogued a palm-leaf manuscript of a hundred and fourteen leaves. He described it in four sentences. The character was transitional Gupta, perhaps a century older than a Parameshvara tantra that Cecil Bendall had catalogued at Cambridge and that carried a date of 859. The first half of the work was called Srauta and the second half Guhya. The leaves had come to him in disorder; he had put them back in order himself, and the whole ran to four and a half thousand verses. Then he moved on. His anchor turned out to be out by thirty-one years. The editors of 2015 adopt Sanderson’s redating of that Cambridge manuscript to 828, first published by Goodall in 1998, which on Sastri’s own reasoning would push the Nisvasa codex back to around 728. They do not go there. They think the two hands similar, and that if either manuscript is the later one it is the Nisvasa, and they close the question by reiterating Sanderson’s judgement that the codex can be assigned approximately to the period 850 to 900.

The painted wooden covers were described much later, in the UNESCO nomination of 2012. Nine deity figures on a red ground inside the front board. Ten fully visible inside the back, and two more only partly visible.

Then the book went back on its shelf, and almost nobody read it for a century.

It is now generally held to be the oldest surviving scripture of tantric Shaivism. A critical edition of three of its five books appeared in 2015, seven years after its editors said it would be ready. The longest book has still not been published. And the case for its antiquity rests on a strange kind of evidence: on the things it has not got.

The counts that fall short

In January 2007, for two weeks, scholars of Shaivism gathered at the Pondicherry centre of the École française d’Extrême-Orient and read the text together. The edition that eventually came out of it names about twenty of them. The report Dominic Goodall and Harunaga Isaacson wrote afterwards, three pages of a project newsletter, contains the sentence that has carried the Nisvasa ever since.

“Many of the lists of theologemes or other entities fall short of expectations formed by reading other Saiva literature: thus, in the Nisvasa, we encounter only 4 kalas where we expect 5, thirty-two tattvas where we expect thirty-six, 2 nadis where we expect minimally 3, and so forth.”

Each of those is a count that later Shaivism fixes and this text has not fixed yet. The tattvas, the constitutive principles of the cosmos, run to thirty-six in the mature system; the Nisvasa gives nine in the Uttarasutra and thirty-two in the Nayasutra, where Shiva at the summit is “the thirty-second, without parts” (dvatrmsakah sivo jneyo niskalah paramesvarah, Nayasutra 4.98, in the reading the editors keep from the palm-leaf against the tidier dvatrimsakah of the copies K and W). The kalas, which Gavin Flood glosses as “the powers that pervade divisions of the cosmos”, number five later; here there are four. The editors suggest they were originally conceived as four Shaktis of the Lord, or as four divisions of the Lord’s Shakti, and that the making of a set of five by the addition of shantyatita came later. Which is the shape of the whole argument in miniature: a group that has not settled how many of itself there are.

The “and so forth” covers more than the report says. There is one initiatory path, the hierarchy of worlds, where later Shaivism has the famous six. There is no graded hierarchy of initiates. Acaryas exist, and there is a short rite for consecrating one at Mulasutra 8.10 to 18, but everyone in the religion appears to be called simply a sadhaka, and the fourfold ladder of samayin, putraka, sadhaka and acarya that later Shaivism runs on is missing. The editors find that ladder already in the Svayambhuvasutrasangraha, the scripture Sadyojyotih commented on around 675 to 725, which means, they say, that the categories were already in existence in the seventh century. Then they decline to bank the point: whether the ranks were widespread enough by then for their absence here to prove anything is, they write, impossible to determine with certainty. There is no seedless initiation for those too busy or too incapable for daily ritual, which in later texts means women, kings and simpletons.

And there is no mala. The innate impurity that clings to every soul, the thing the mature Shaiva Siddhanta exists to diagnose and its initiation exists to destroy, is never mentioned in the scripture that school counts as its own. The editors are careful about what that proves. They noticed in 2007 that the same silence runs through several other early Siddhanta scriptures which look, on other grounds, later than the oldest parts of the Nisvasa: the non-eclectic recensions of the Kalottara, the surviving chapters of the Rauravasutrasangraha and of the Nisvasakarika, and the Sarvajnanottara. They add one qualification in a footnote, which is the sort of thing that tells you how they work: the Sarvajnanottara does use the word mala, but of the retributive force of past actions, not of innate impurity. What the absence marks is a stratum, not a single book.

The inference runs: a system still finding its numbers is a system near its beginning. What makes the argument stronger than it sounds is where the missing thirty-six finally turn up. They arrive in the Diksottara, a text added to this same corpus at a later date, so the full count does not arrive from outside it. One caveat is worth keeping. The Diksottara is an addendum to the Nisvasakarika, itself an appendix of uncertain date, and it survives in late South Indian paper transcripts rather than in the Nepalese codex. The sequence can be watched assembling itself. It cannot be watched inside one manuscript.

Five books, and one of them is a doorway

The corpus describes its own shape. At the head of its fourth book it says that three have been taught in order, the Mula, the Uttara and the Naya; that the Guhyasutra is the fourth; and that joined with these it becomes one samhita, complete, under the name Nisvasa. Then the opening half of the next verse adds that joined with the Mukha it is the Nisvasasamhita, the second half says that the five streams are proclaimed by the Mukha, and a further half-verse says that joined with that it becomes full. The editors note, drily, that these four and a half verses may themselves have been added later to explain the unity that resulted when all five books came to be copied together.

The order of composition is argued from backward references. The Uttarasutra refers to the sixth chapter of the Mulasutra; the Guhyasutra, mentioning rosaries, says they were described earlier, and they were, at Uttarasutra 4.24. The editors are unusually direct about the result: they are confident the four sutras were composed in exactly the order the manuscript transmits them. The fifth is another matter. The Nisvasamukha stands first in the codex and they take it, provisionally, for the latest layer. Being an introduction, it must presuppose at least one of the books it introduces, but since it never discusses their contents there is no telling whether all of them existed when it was written, and they say so. There is one allusion: Nisvasamukha 3.28 mentions the properties of the water of Kedara and appears to point at the vidya-mantra, which suggests knowledge of Guhyasutra 16 and 17. They add, at once, that both texts might be drawing independently on earlier material. Then they rest on neither argument. What they write is that at the moment they are provisionally assuming the Nisvasamukha to be the latest layer, which is an honest way of saying that the question is open.

The fifth book is the odd one. The Nisvasamukha is not a tantric text. It is a preface, and what it prefaces is the religious landscape out of which the Mantramarga came. It divides religious literature into five kinds: the worldly, the Vedic, the philosophical, the atimarga and the mantramarga. Three of its four chapters are given to the first of those, the lay Shaiva religion of wells and ponds and gardens and pilgrimage and fasting. Nirajan Kafle, who edited it, points out that no other canonical Shaiva scripture before the tenth century opens with non-tantric content, and reads it as a book written to introduce the Mantramarga to everybody else.

The road out of the Atimarga

This is the frame the Nisvasa makes visible, and it is worth stating plainly because the handbooks tend to start after it.

Before tantric Shaivism there was the Atimarga, the path beyond: the ascetic Shaiva orders, the Pashupatas and the groups the sources call Lakula, offering liberation and nothing else, to brahmin men. The Mantramarga, the way of mantras, is what the Agamas and tantras teach, and the editors characterise the shift precisely. It placed strong and perhaps equal emphasis on two goals, liberation and the enjoyment of supernatural powers. It made initiation a transformative rite rather than a door. And it offered that liberation, in their words, “not only to brahmin males, as appears to have been the case in the Atimarga, but to those of all varnas and, in some cases, to both sexes.”

The Nisvasa is the text in which that shift is still happening. Long stretches of the Nisvasamukha and of chapters three to seven of the Guhyasutra carry the Lakula ladder of worlds, borrowed from an Atimarga source that is now lost. A section of the Nisvasamukha, which Acri gives as 4.69 to 97 and Kafle’s published edition as 4.70 to 88, is for the most part a paraphrase of the Pasupatasutra. Andrea Acri, who edited a passage of the Guhyasutra that classifies the Atimarga groups from the new tradition’s point of view, puts it this way: the elements of continuity indicate that the shift was still in the making, and the text was composed when the two traditions shared much and were only beginning to compete.

Dated by the shape of letters

The most quoted piece of evidence for the date is also the strangest, and it is weaker than its reputation.

The first chapter of the Nayasutra teaches the practitioner to shape his own body into the graphs of each letter of the Sanskrit syllabary. Not the sounds. The letter-forms. Which means the text presupposes a particular set of shapes, and letter-shapes can be dated. The workshop’s report says: “we agreed that the script presupposed appeared to be North Indian of between the 5th and 7th centuries AD.”

We agreed. No palaeographer is named. No comparandum, no inscription, no footnote. It is a recorded show of hands in a conference report.

The published edition does better. It names what it compared the letters against, Somdev Vasudeva’s study of the alphabet-goddess Malini and Georg Bühler’s plates of northern alphabets from about 350 to 800, and it prints the evidence for the reader to look at: photographs of the letter ja in the Cambridge Parameshvara manuscript and in the Nisvasa codex itself, and a whole Gupta font that Somdev Vasudeva built from Sander’s tables. It also moves the lower bound, giving the bracket as the fourth to seventh centuries where the newsletter had the fifth. There is still a circularity worth noticing, which neither document addresses: the script is inferred from a verbal description of postures imitating letters, and then used to date the text that describes them.

The palaeography is one strand of six. The editors name them: palaeography, iconography, terminology, theology, social religion and intertextuality, and they say so explicitly, that individually each piece may seem merely suggestive and it is the cumulative weight they find compelling. The iconographic strand is a good one. The five-faced Sadashiva of the later tradition has Tatpurusha in the east almost without exception. In the Nisvasa, at Nayasutra 3.42 to 44a, the eastern face is Ardhanarishvara, the half-woman form, and it belongs to a four-faced Ishvara rather than to the five-faced Sadashiva of the later tradition, which is itself part of the point.

What it handed on

The Nisvasa was read, and the reading left marks.

More than a thousand of its verses were taken up and adapted into the Svacchandatantra, the scripture that became the basis of the majority Shaiva cult of medieval Kashmir and the one Kshemaraja chose to comment on in the eleventh century. The Svacchanda was in turn cannibalised, in much the same way, by the Tantrasadbhava. That is a demonstrated chain of transmission running across five centuries, and the editors lead with it rather than with the date. It is worth noticing which evidence the people who know the text best put first.

It carries something else, at Uttarasutra 1.23 and following, with the names themselves running from 1.26 to 1.30 and the total stated in 1.31: what is probably the earliest surviving list of a canon of twenty-eight scriptures, with its own name on the list. That list, with small differences, is the one now known as the canon of the Shaiva Siddhanta. Which is the awkward part. The text that transmits the charter shows no sign of knowing that the school exists. Its editors read the list as a record of works, some or even most of them perhaps mythical, belonging to an undifferentiated Mantramarga. The school arrived afterwards and found its canon already drawn up.

How that canon was later divided, and into how many seats, became an argument of its own, and the sources do not draw the diagram the handbooks print.

The figure, and how it travelled

What the workshop actually wrote was this:

“On the basis of our reading so far, we are inclined to place the earlier parts of the text between 450-550 AD.”

Four hedges in twenty-two words. On the basis of our reading so far. We are inclined. The earlier parts of the text, which is not the text and is not a named book. And a collective we, with the sentence immediately after it recording that no final conclusion on the layering had been reached.

Follow the sentence outward and it hardens at every step. Alexis Sanderson, in 2009, narrows it to a named book and keeps the hedge: the Mulasutra was probably composed between 450 and 550. The editors do the same narrowing in 2015, writing of the Mulasutra at c. 450 to 550 AD. Acri, in 2014, attaches the same range to the Nayasutra, citing the report that says the Mulasutra. The UNESCO Memory of the World nomination, which inscribed the manuscript in 2012, drops the hedges: the earliest part of the text was written between 450 and 550. The English Wikipedia article on Tantra, which is where most readers meet this text, gives it a bare century with no range and no stratum, hedged only by a passive, and footnoted to a page of a general encyclopedia survey by a Buddhologist. The same sentence files the Nisvasa under Shaiva Siddhanta, which is where tradition puts it and where its editors put it too. The point worth making is narrower and more interesting than a contradiction: the text itself never says so, and shows no sign of knowing that such a school exists.

And the question mark went with them. The workshop itself was called Early Shaivism: the Testimony of the Nisvasatattvasamhita. It was the report on it that asked, in its title, “The Earliest Surviving Saiva Tantra?” The 2015 book states it flat in its title, on the cover and on the title page both. Inside, the editors write “arguably” on one page and “may be” on another, and the publisher’s blurb narrows the claim itself, to “probably the oldest surviving complete scripture of the Mantramarga.” The hedge survives everywhere except in the title.

The objection that was never withdrawn

In 2014, in two footnotes of The World of the Skandapurana, Hans Bakker said he did not believe it.

“Since this dating is mainly based on palaeography, and developments therein are difficult to pin-point within an interval of a hundred years (cf. Sanderson 2013, 221: ‘notoriously unreliable’), we have serious doubts about this early date and find it more likely that the corpus began to evolve one century later, that is more or less simultaneously with the Skandapurana.”

He repeats it a hundred and forty-one pages later in six: the early date rests, in his view, on insufficient grounds. He offers no counter-evidence of his own, and he is aiming at the pre-edition statements rather than the 2015 introduction, which had not yet appeared.

The editors answered him, in a footnote. One sentence. They quote four words of him, that the dating is “mainly based on palaeography”, and say they hope the discussion that follows will go some way towards dispelling his scepticism. What follows is the six-strand argument. It is a reply, and it is the only one on the record.

Then he reviewed the finished edition, nine pages of it, in the Indo-Iranian Journal in 2018. This article has not read that review. It is paywalled, and whether he there held his ground, softened or gave it up is a gap in what is written here rather than a gap in the record. What can be said is that Kafle’s 2020 volume, the most recent work from the editors’ circle, notes the 2014 objection in two sentences, declines to adjudicate it, and does not cite the review at all.

One verdict from after the edition appeared is readable. Gavin Flood, reviewing the same book for the Journal of the American Oriental Society in that same year, wrote that the editors present thorough evidence for their reasoning about the range of dating for the different layers of the text. That is one reviewer, and he is not the one who objected. But it is on the record, and it is worth more than the silence this article would otherwise leave in its place.

So the position is not that the date is settled. It is that the editors’ date is the standard one, that an objection to it was published in 2014 and has not been withdrawn in anything readable here, and that a specialist reviewing the finished book found the evidence for it thorough.

A spell it shares with the Buddhists

Here is the thing least likely to appear in a popular account.

The Nisvasa and the Buddhist Manjusriyamulakalpa contain the same recipe, almost word for word. An effigy is made of salt, smeared with blood in the Shaiva version, cut apart with a sharp blade beginning at the feet, and a thousand and eight oblations are offered, after which the person represented comes under the practitioner’s control. In the Nisvasa it stands at Guhyasutra 14.65. A third text has it too, the Vinasikhatantra, which is the other candidate for the oldest surviving tantra.

They also share a way of grading attainment. A recipe is performed, and what it yields is announced by a sign: if there is heat, the power to cover great distances on foot; if smoke, the power to disappear; if flames, the power to fly. That is Guhyasutra 10.31 to 32. Goodall and Isaacson note that the same three levels occur in Buddhist tantric works, the Amoghapasakalparaja among them, and then add the sentence that matters: “the classification is extremely rare in Saiva literature outside the Nisvasa, and we have so far found it only in two places.” The two are a chapter of the Brahmayamala and a passage of the Sarvajnanottara. In 2007 they had named only the first.

Read that slowly. The scheme is at home in Buddhist tantras and nearly absent from Shaiva ones, except here, in the text used to anchor Shaiva priority. It does not settle anything about who borrowed from whom, and the long argument about the direction of borrowing is fought over other texts entirely, the Yoginitantras of several centuries later. But it is a reminder that these traditions shared a world and a ritual vocabulary before they were rivals with a literature to defend.

Who it was for

The religion of the Nisvasa is open and hierarchical at the same time, and it does not appear to notice the tension.

Initiation is available to all four varnas. The waiting period is not the same for all of them. The Mulasutra sets it out at 8.6: a brahmin after three, four or five years, a Kshatriya after six, seven or eight, a Vaishya after ten or twelve, and a Shudra after twenty. That is a long probation, and where it sits among its neighbours depends which neighbour you pick. The Matasara asks twelve years of a Shudra, which is less. The Nepalese manuscript of the Sarvajnanottara says instead, of a Shudra, that one should not have the knowledge given at all.

Even the ground is caste-coded. The earth of the sacrificial site should be white, red, yellow or black according to the varna of the worshipper, which is the old pun on varna as both colour and class, turned into a rule about where to dig.

And then, in the yogic register, the whole thing dissolves. At Nayasutra 4.56 to 57b the adept who has arrived sees that enemy and friend are equal, and that the brahmin and the dog-cooker are equal, and knowing everything to be Shiva he becomes a Siddha. The teaching that abolishes the distinction is the same teaching the Uttarasutra says to give a Shudra only if he is obedient.

What is missing from the sutras is most of social religion, and it matters which absences count. Nobody would look for birth or marriage rites in a Shaiva scripture. Death rites are the real silence: later Shaiva scriptures treat them at length, and here there is one passing half-verse, at Guhyasutra 11.38. No festivals. No monasteries. The editors draw the conclusion carefully: this looks less like an institutionalised religion than a young movement that saw itself as a path beyond established social religion, and it is, in their word, parasitic on the brahmanical substrate that supplied everything it does not teach. The opening to householders does happen in this corpus. It happens in the Nisvasamukha, the lay preface, and not in the books of initiation.

A stone at Angkor

In February 1922 Henri Marchal found two laterite doorjambs lying inside Prasat Banteay Kdei at Angkor, re-used, inscribed on the reverse of their carved faces. Forty-five Sanskrit verses praising a guru and his pupil Hrisikesa, a Shaiva priest of a Vaishnava family who served four kings.

At verse 36 the inscription says that at his consecration Hrisikesa received the naisvasamandali diksa. Sanderson reads that as initiation into the mandala taught in the Nisvasa. Two verses later the stone says he had mastered all the samhitas, beginning with the Sarvajnanottara, which is unmistakably a scriptural register.

The inscription carries no date. It is placed in the reign of Rajendravarman, 944 to 968, from the accession date given in verse 39 and from the four kings the man served. Louis Finot published it in 1925 without connecting Naisvasamandali to any book, reading it as the name of an initiation. The identification with the Nisvasa is Sanderson’s, from 2001, it rests on the name, and it has been taken up rather than argued with.

Taken for what it is, it says that by the middle of the tenth century a scripture called the Nisvasa was known fifteen hundred miles from Kathmandu, held canonical, and operative at the highest level of Khmer royal religion. It is a fact about reach and prestige. It is not an earlier date. The Nepalese manuscript is older and demonstrably contains the text.

What survives

The manuscript is damaged at the edges, and the edges are where the letters went first. Three twentieth-century copies exist, made when more of those letters were still legible, and the editors use them to recover what the original has lost. One of them was written out in Kathmandu in 1912 for Dr Paira Mall, a collecting agent, and it now sits in a medical-history library in London. Its colophon says the copyist was a Nepal-dwelling bauddhasevita vajracarya. The Wellcome catalogue takes that for his name. The editors take it for a description, and read it as telling us only that he was a Buddhist Vajracarya.

More of the text was recovered a different way. Two scholars at the French Institute of Pondicherry, Sambandhasivacarya and Anil Kumar Acharya, noticed that a later work, the Sivadharmasangraha, had borrowed several hundred verses of the Nisvasamukha and adapted them. The borrowing could be run backward. The younger text repaired the older one, and one of the two men who saw it carries a hereditary title of the South Indian temple priesthood.

The Nisvasa has no living claimants. No lineage recites it, no school teaches it, no guru invokes it. It is better off than the Saura tantras, which left eighty-five titles and no text at all, but only just. Its one piece of public standing is the UNESCO register entry of 2012. The nomination is a careful document in places, quoting Sastri, Bagchi, Sanderson, Goodall and Isaacson by page, and Kafle’s then unpublished thesis at length. It is where the hedges go to die. Its summary calls the object the earliest surviving tantric manuscript, which is a claim about a physical thing and stronger than any claim the scholars make about the text. It states that the earliest part was written between 450 and 550, flat. And it tells the reader that tantrism influenced Islam in India.

The editors themselves are more careful than their own title page. In the introduction to the 2015 volume they write that by calling the Mulasutra possibly the earliest surviving Shaiva tantra they do not mean it was the earliest Shaiva tantra, and that its concision, indeed its abruptness, rather suggests it was not the only text of its kind when it was written. Elsewhere they name a book of comparable antiquity, the Vinasikhatantra of the Vama current. And then they say, carefully, that little makes it possible to declare anything certain about which of the two is older, before pointing to a couple of surprising features that may be pointers to the Nisvasa’s greater antiquity.

A scripture that does not use the word tantra the way we use it. That does not know the name of the school which claims it. That counts four where the tradition counts five, and leaves out the impurity the tradition was built to explain.

The oldest thing we have is the one that knows the least.