Governed intelligence · for regulated banks, NBFCs and the firms that own them
When your board asks if the AI can be switched off, show them. Don't tell them.
New model-risk rules mean your regulator now expects you to prove your AI can be stopped, and that nothing it touched was left behind. Amzaa answers that on a screen: one action stops every agent, a deterministic floor keeps your compliance running with the AI off, and a sealed trail proves it, a trail your regulator can verify without trusting us.
The floor still runs. Deterministic checks, no model, no prompt. It has no switch, because a floor with a switch is not a floor.
1 switch
One kill-switch, one audit point. Stop every agent in one action, and every action it ever took is sealed in one place.
Sealed
A hash-chained, Merkle-sealed trail. Its root is published outside the database, to you. Tamper with any row and verification fails.
Why people come to us
Two reasons. One is on your desk today. One is coming for everyone.
On your desk now
"My regulator is asking about AI, and I cannot prove we control it."
A model-risk framework lands and the board turns to you. Can the AI be switched off? Can you show nothing leaked? Can an examiner check it themselves? Right now the honest answer is a slide deck and a promise. Amzaa makes it a live demonstration, and a sealed record they can verify.
Coming for everyone
"We are putting AI agents across the business, and nothing governs them."
Within a year every function will run agents that read, decide and act. The question stops being whether you have AI and becomes whether you can govern it: stop it, prove it, show who approved what. Amzaa is the rail those agents run on, built before you needed it.
Every feature is here because it answers a question someone will ask you.
The board · the regulator
"Turn off the AI. Now prove nothing it did is still running."
One kill-switch
One action stops every agent at once, and fails closed if it cannot confirm. Count the AI's actions before and after. The answer is a number on a screen, not an assurance.
The board said no to AI
"We are not allowed to run the model. Does the platform still work?"
The deterministic floor
Your compliance checks run with no model at all: coverage gaps, broken lineage, stale evidence, found the same way every time. It ran last night with the AI switched off, and has no off switch of its own.
The auditor · the examiner
"Who changed this record, and how do I know your log wasn't edited?"
A sealed, verifiable trail
Every write records who, when, what and why, sealed in a hash chain whose root you hold outside our database. Tamper with any row and verification fails. They check it themselves.
The CFO
"Why are we paying for nine systems and five consultants to run governance?"
One engine, configured
Audit, risk, policy, controls, compliance, vendor risk, issues and submissions on one engine. A new solution is a configuration, not a six-month project. Millions a year becomes a fraction of it.
The PE firm · the group
"I have thirty portfolio companies. I need to govern all of them from one place."
One graph, many entities
Every company runs its own governance and rolls up into one view. Each tenant's sealed trail is independent, so one company's records can never appear in another's. Sell a company and its governance travels with it.
The demo we open with
Tamper with a sealed record. Watch the chain break.
This is the moment that ends the meeting. Every write is hash-chained to the one before it. Change a value in a sealed row, and re-verifying no longer matches, from that row all the way down. Shown here as an illustration of the behaviour.
chain verified · 5 of 5 rows intact · root matches your published copy
In the product the trail is a Merkle-sealed, write-once table, and its root is signed and published to you, so even we cannot rewrite it without your copy disagreeing. See how the proof works →
A floor that can be switched off is not a floor.
The deterministic floor
The AI was off all night. It still caught everything.
The floor checks coverage, orphaned citations, broken lineage and stale evidence with no model, no prompt and no inference. The same inputs give the same findings every run. There is nothing to hallucinate and nothing to explain to a regulator.
It runs on a schedule by itself, it ran last night with the AI switched off, and it has no off switch. That is the sentence most platforms cannot say.
These are not features bolted onto a governance tool. They are the foundation it stands on.
AI-off proof
Deterministic to the core
The floor never used a model, so switching the AI off does not degrade it. A platform built model-first cannot remove the model and keep working. That is a rebuild, not a setting.
Sealed by design
The chain is how writes happen
Tamper-evidence is not a log you switch on. Every write is derived and sealed by the database itself. There is no path to change a record that skips it.
Patented
The strongest claims, filed early
Time-travel, legal hold, a Merkle chain that breaks on a touch, and a regulator replay sit behind a patent portfolio, including a kill-switch cluster filed before the RBI's 2026 draft framework.
Design partner programme
Bring us the question your regulator is going to ask.
A small cohort of regulated banks, NBFCs and the firms that own them. Early access, real influence, pricing that holds.
We are pre-launch and we will not dress it up. There are no logos on this page because there are none to show. Come and try to break the chain.
Thanks — it's with us.
We read every one of these ourselves. Here's what you sent: