How is Manthan different from a frontier model on a public cloud?

Manthan runs inside your perimeter — your cloud, your keys, your jurisdiction — so your data never leaves the building, and every call is masked, shielded, and recorded as a signed Trust Receipt for audit. It is the reasoning tier without the data-exfiltration trade.

Which languages does Manthan handle?

Indian languages and English, including code-mixed text — reasoning across the languages your organisation and its customers actually use, not English-only with translation bolted on.

Can we build on Manthan now?

Yes — Manthan is in early access. Teams can build against it today and help shape it toward general availability, which is targeted for October 2026. Talk to us to get access.

MODEL · GENERAL REASONING

Early access · GA Oct 2026

Manthan

मंथन

Reasoning that stays inside your walls.

Manthan is AgentAnywhere's sovereign general-reasoning model — analysis, synthesis, long-context understanding, and structured output across enterprise work, strong in Indic languages and English. It runs inside your perimeter, behind the same trust layer as the rest of the platform, with a signed Trust Receipt on every call. In early access now.

Deployment: In your perimeter — region-pinned or air-gapped. Your keys, your jurisdiction.
Languages: Strong across Indic languages and English, including code-mixed input.
Every call: PII masked, output shielded, and closed with a signed Trust Receipt.
Proven: Earns its place through an execution-graded gate — we publish the numbers.

The reasoning tier, without the exfiltration.

Most enterprises reach for a frontier model the moment a task needs real reasoning — and in doing so, send their most sensitive documents to someone else's cloud. For a bank, a hospital, or a government, that trade is not available. The work that most needs a reasoning model is the work that most cannot leave the building.

Manthan is the answer to that bind: a capable general-reasoning model that runs where your data already lives, under your governance, with proof on every call. You get the reasoning tier without giving up the perimeter.

What Manthan does.

A general-reasoning model for the analytical work that fills a knowledge worker's day.

Long-context understanding

Reads across long documents and multi-part inputs and holds the thread — so the answer reflects the whole, not the last page.

Analysis & synthesis

Pulls signal from dense material — reports, policies, research — and synthesises it into something a person can act on.

Structured output

Returns answers in the shape your systems expect — fields, tables, and typed records — not just prose to re-parse.

Indic and English

Reasons across Indian languages and English, including the code-mixed way people actually write, rather than English-only.

Drafting & decision support

Drafts, summarises, and lays out options with their trade-offs — support for a human decision, with the human in the loop.

Answerable, not just answered

Every call closes with a signed Trust Receipt, so a reasoning step in a regulated process is auditable after the fact.

Behind the same trust layer as everything else.

Manthan runs behind the platform's trust layer. Veil masks PII before the model sees it, Kavach shields each request and response against prompt injection and unsafe output, and every call closes with a signed Trust Receipt.

It deploys inside your perimeter, on your keys and jurisdiction: SOC 2 and ISO 27001 certified, independently assessed for HIPAA and GDPR, designed to support India's DPDP Act and the RBI FREE-AI framework. The reasoning happens where the data is — not the other way around.

Who Manthan is for.

Enterprises and institutions that need a capable reasoning model for analysis, drafting, and decision support — but whose data cannot go to a public frontier model. Banks, insurers, hospitals, governments, and the teams building sovereign applications on top of the platform.

Where Manthan is today — stated plainly.

Manthan is in early access, with general availability targeted for October 2026. It is further along than a preview, and we still say exactly where it is rather than rounding up.

Access is open to teams who want to build against it now and help shape it toward GA. As with every model here, Manthan earns its place through an execution-graded gate, and we publish the numbers when they are real.

FAQ

Frequently asked questions.

What is Manthan?: Manthan is AgentAnywhere's sovereign general-reasoning model — analysis, synthesis, long-context understanding, and structured output, strong in Indic languages and English. It runs inside your perimeter behind the platform's trust layer, with a signed Trust Receipt on every call. It is in early access, with general availability targeted for October 2026.
How is Manthan different from a frontier model on a public cloud?: Manthan runs inside your perimeter — your cloud, your keys, your jurisdiction — so your data never leaves the building, and every call is masked, shielded, and recorded as a signed Trust Receipt for audit. It is the reasoning tier without the data-exfiltration trade.
Which languages does Manthan handle?: Indian languages and English, including code-mixed text — reasoning across the languages your organisation and its customers actually use, not English-only with translation bolted on.
Can we build on Manthan now?: Yes — Manthan is in early access. Teams can build against it today and help shape it toward general availability, which is targeted for October 2026. Talk to us to get access.

Put the reasoning tier inside your perimeter.

Manthan is in early access, GA targeted October 2026. If you need real reasoning but your data cannot leave the building, talk to us about access.

Talk to us about Manthan