For boards and finance leaders

The compliance clock is a date, not a forecast.

The Digital Personal Data Protection Rules 2025 were notified on 14 November 2025 with an eighteen-month phased compliance period. It ends on 13 May 2027. The largest penalty in the Act is not attached to a form somebody forgot to file — it is attached to the safeguards themselves, which is the category access control sits in.

13 May 2027: End of the DPDP Rules’ eighteen-month phased compliance period, counted from notification on 14 November 2025.
Up to ₹250 crore: The Act’s highest penalty: a Data Fiduciary’s failure to maintain reasonable security safeguards. Every other category tops out lower.
180 days, in India: CERT-In requires logs of all ICT systems to be retained for a rolling 180 days and maintained within Indian jurisdiction.

What we will not tell you is what proportion of a breach cost this product removes. Nobody can calculate that honestly, and a vendor who hands you that number should be asked how they derived it. What we will tell you is which line the control attaches to: it does not stop the intrusion. It changes what an intruder can do with what they take, and how long they keep it. When the identity plane never leaves your data centre, that stops being a project and becomes a property of where you installed it.

It does not make your organisation compliant. No product can do that, and any vendor who says otherwise is telling you something checkable and false.

Take a live token and try to reuse it

Sources: Press Information Bureau, Government of India, ‘DPDP Rules, 2025 Notified’; the Digital Personal Data Protection Act 2023; CERT-In Directions No. 20(3)/2022 of 28 April 2022, under section 70B(6) of the Information Technology Act 2000.

The platform

One system where a stack of vendors used to be.

Okta and Zscaler and a token vendor do, between them, roughly what Pramaan ID does on its own — except they do it in someone else's cloud, in someone else's jurisdiction, over your users, your device inventory and your access logs.

Shipping

Pramaan ID

Written almost entirely in Rust, with one shared security core — sso-core — compiled identically into the server, the desktop client and the phone. One audited implementation of the security-critical code, everywhere it runs.

Enterprise identity

OIDC/OAuth with PKCE and rotating refresh tokens, a native SAML 2.0 IdP, SCIM provisioning, an LDAP server for the apps that still need one, directory sync from Active Directory or FreeIPA, passkeys, and a self-service account portal.
Zero-Trust access

Per-application ZTNA tunnels that fail closed, a forward-auth proxy, browser-delivered RDP/SSH/VNC, a secure web gateway with private DNS filtering and selective TLS inspection feeding inline DLP.
Trust and proof

Device-bound tokens (DPoP, RFC 9449) tied to a key that never leaves the Secure Enclave or TPM, post-quantum co-signed approvals, device attestation, live revocation over CAEP/SSF, and an RFC 6962 transparency log whose inclusion proofs anyone can verify.
Built for AI

A first-class agent registry, an MCP tool-broker that decides per tool call through the same policy engine as everything else, and signed action receipts you can verify offline against the public key.

Boundary The AI is never on the hot path. If the model disappears, logins keep working — administrators simply lose an assistant.

Why it exists

Modern identity is a paradox for a sovereign operator.

Sovereignty and modern security are sold as alternatives. Three walls make that trade-off feel inevitable.

The best tools are cloud-only

Okta, Auth0, Entra, Zscaler, Cloudflare — every one of them requires your directory, your device inventory and your audit trail to live somewhere you do not control and cannot inspect.
The on-prem alternatives are stale

Self-hosted identity gives you sovereignty and then leaves you years behind on passkeys, device posture, ZTNA and everything that has happened since.
“Zero Trust” is mostly slideware

Vendors say it; few prove it. Access is still granted on a bearer token any thief can replay, from any device, in any state of health.

What makes it different

Six decisions, each of which costs something.

None of these are free. They are the trade-offs a platform makes when it refuses to choose between owning your identity plane and running the current state of the art.

01
Your trust roots, your infrastructure

Every component runs on your infrastructure, and the trust roots are yours: the device CA is generated on your own broker, never issued by us. The model is self-hosted. There is no licence server, no telemetry and no outbound dependency anywhere on the login path — and the identity plane installs with no network egress at all: images move on disk, TLS comes from your own internal CA, and the signing key is generated on your host. We prove that on a network with no route out, against a control, rather than asserting it.
02
Device-bound everything

A stolen token is worthless. Tokens bind to a hardware-class key that never leaves the Secure Enclave or TPM; approvals are signed by that key and co-signed with ML-DSA (FIPS 204) against harvest-now-decrypt-later.
03
AI proposes, an engine decides

The Policy Copilot drafts a rule and points at the existing rules it drew on, rather than inventing a justification. It declines to answer when it is not confident. A deterministic engine then checks every draft against the real policy model before a human is asked to approve it — the assistant proposes, it never decides.
04
The log records

Every verified decision writes a reproducibility receipt — the model's SHA-256, the prompt hash and the grammar hash — so any decision can be replayed and audited long after the fact.
05
Posture you can see and fix

Twelve device signals — disk encryption, firewall, screen lock, patch level, automatic updates, EDR, Secure Boot, TPM, central management, SIP and Gatekeeper, and a minimum OS floor — each weighted and admin-tunable, each collected silently, each with grounded remediation steps for the user's actual platform.
06
One Rust core, everywhere

Sign-in, token validation, device signatures and posture scoring live in one audited Rust crate, compiled into the server, the desktop client and the phone. A fix to security-critical code lands in all three at once, instead of being written three times and drifting.

See it work

Copy the pass. It still will not let you in.

Once you have signed in, almost every system hands your browser a pass — and for most of them, anyone holding a copy of that pass is you. Lift it off a laptop, a log or a proxy and it works from anywhere in the world. Pramaan’s pass is tied to the one device it was issued to, and that is re-checked on every request.

One access token used two ways

200

Let in

The device it was issued to

It still holds the key the token was stamped with, so it can prove the token is its own — every single request.

401

Refused

Any other machine, same token

It has a perfectly good key of its own. It simply is not the key, and the token names only one.

Nothing is cancelled and nothing runs out. Seconds after the refusal, the original device presents the very same token and is let straight in. The token was never the secret — the device is.

Run the replay yourself How it works

The lab signs you in with a demonstration account, generates a key inside your browser that cannot be exported, issues a token bound to it, and then invites you to steal that token and use it somewhere else. Device binding is opportunistic by default, so a fleet can be moved onto proofs before they are demanded; the demonstration deployment demands them.

Identity for AI agents

An autonomous agent is a new kind of principal. Treat it like one.

An agent registers against exactly one human owner, receives a least-privilege scope ceiling it can never exceed, and is issued short-lived, device-bound tokens — so a leaked agent secret is worthless off the owner's enrolled machine.

Every action is provable

The broker returns a signed action receipt: an offline-verifiable proof, checkable against the public key with no trust in the server, of exactly what the agent was allowed to do, on whose behalf, and when.
Per-tool decisions

A Model Context Protocol tool-broker makes Allow/Deny calls for agents through the same policy engine that governs every human surface — and fails closed.
One kill switch

Disable an agent instantly. Every token it ever minted and every tool it was authorised to call is already in the transparency log.

Against a cloud SaaS stack

Not a cheaper SaaS. A different place to stand.

The differences that matter are not features. They are where the data sits, who can prove what, and what a stolen credential is worth.

Concern

Pramaan ID

Cloud SaaS stack

Where identity data lives

Your datacenter. Air-gappable.

The vendor's cloud, under the vendor's jurisdiction.

Stolen access token

Useless — bound to a hardware key on one device.

Replayable from anywhere until it expires.

Policy assistance

On-prem model, deterministically verified, logged.

Vendor-cloud model, or nothing.

Audit evidence

RFC 6962 log — you verify the proofs yourself.

A vendor-attested export you have to believe.

AI agent identity

First-class principal, scoped and kill-switchable.

A service account with a long-lived secret.

Architecture

One core, one authority, one boundary.

Four tiers and one process. Every request — a browser session, a tunnel, a tool call from an agent — is decided by the same policy engine against the same evidence, and nothing is cached across a change in posture. There is no control plane somewhere else to be unavailable.

companion-broker: Rust · axum. The IdP and SASE core: OIDC, SAML, ZTNA policy, secure web gateway, LDAP server, SCIM, device CA, agent registry, transparency log, audit.
sso-core: The shared security crate. Identical crypto in the server, the desktop client and the phone — the handset and the server produce byte-identical signatures, and interop tests prove it on every build.
Pramaan app: Flutter and Rust. Push approvals, passkeys, TOTP, biometrics, on-device posture collection.
pramaan-desktop: Rust. ZTNA tray client, pramaanctl, silent posture agent, minisign-verified auto-update.
admin-ui: Svelte 5. Users, RBAC, policy, posture weighting, audit.
account-ui: SvelteKit. Sessions, devices, sign-in history, self-service recovery.

Pricing & packaging

Four tiers, one platform — priced to consolidate.

Named-user bands, self-declared — there is no licence server to call home to, and no per-agent metering. Annual subscription, or perpetual with annual maintenance. Baseline security — SSO, MFA, passkeys and the tamper-evident audit log — is in every tier.

Tier 1

Pramaan Identity

Replace a legacy IdP with a sovereign, on-prem one.

from ₹1,500 / user / yr · US$35 · ex-GST

OIDC / PKCE & SAML 2.0 (IdP + SP)
Passkeys, push-to-approve, MFA
Built-in LDAP · SCIM 2.0 in & out
Tamper-evident audit log
Admin & self-service consoles

AI-agent identities included: 25

Tier 2 · Most popular

Pramaan Zero Trust

Add Zero-Trust access to your private apps.

from ₹4,500 / user / yr · US$95 · ex-GST

Everything in Identity, plus:

ZTNA gateway — private apps, no inbound
Device-bound (DPoP) access tokens
Continuous device posture
Secure Web Gateway + DNS / DLP
Per-request, fail-closed access

AI-agent identities included: 250

Tier 3

Pramaan Sovereign

Consolidate IdP + ZTNA + PAM under one plane.

Contact sales

Everything in Zero Trust, plus:

JIT privileged SSH / RDP — no standing creds
Hardware key attestation
First-class AI-agent identity
Post-quantum audit anchoring + SIEM
HA / multi-site

AI-agent identities included: Unlimited

Tier 4 · Enterprise

Pramaan Sovereign+

Government, defence & critical infrastructure.

Contact sales

Everything in Sovereign, plus:

Fully air-gapped installation
Source-code escrow
Sovereign SLA + on-site support
Custom integrations
Make-in-India / CERT-In alignment

AI-agent identities included: Unlimited

Indicative list, per named user / year, ex-GST — prices fall with volume bands (≤500 / ≤2,000 / ≤10,000 / unlimited). Each tier includes one production deployment; non-production is free. Pramaan Sovereign is priced to replace an IdP, a ZTNA gateway and a PAM tool — set it against your current three renewals. Talk to us →

Own your identity plane

Prove every decision, on your infrastructure, on your terms.

A live issuer is answering now, and the documentation covers deployment, the full configuration surface and the integration recipes end to end.

Read the documentation Talk to us

Proof,not trust.

Pramaan ID

Enterprise identity

Zero-Trust access

Trust and proof

Built for AI

The best tools are cloud-only

The on-prem alternatives are stale

“Zero Trust” is mostly slideware

Your trust roots, your infrastructure

Device-bound everything

AI proposes, an engine decides

The log records

Posture you can see and fix

One Rust core, everywhere

Let in

Refused

Every action is provable

Per-tool decisions

One kill switch

Pramaan Identity

Pramaan Zero Trust

Pramaan Sovereign

Pramaan Sovereign+

Prove every decision, on your infrastructure, on your terms.

Proof,
not trust.