The AI proposes. Something separate must decide what is allowed. Okren applies the institution’s limits to every move routed through it before money moves, and keeps a record of every answer.
Phone view shows a short version. Open the full demo on a desktop to see everything.
Illustrative demo with sample data. This is the dashboard your team opens. What you are looking at Open the full demo
Every move gets one of four plain answers.
| Answer | What happens to the move | Example |
|---|---|---|
| Approve | The move is inside what your institution allowed. It goes ahead under the authority you already set. | A $4,200 payment to a known supplier. |
| Reduce | Only the amount inside the limit goes ahead. The rest does not move, and the record shows both amounts. | An agent asks for $25,000. $10,000 goes out. |
| Ask | The move is held for a person you name. Nothing runs until they answer. Waiting is never approval. | A $120,000 wire, over the $50,000 wire limit. |
| Stop | The move does not go. The answer is recorded. | A payment to a recipient that was never approved. |
Examples are illustrative. Your institution sets the amounts and the rules.
Multiple systems, each with its own controls. One independent check across all of them, before money moves.
Different systems do different jobs. Okren is designed to apply the same limits whichever one carries out the move, independent of the AI and of the platform.
AI is missing
independent supervision
These systems can complement Okren. None of those responsibilities, by itself, is independent supervision of the whole proposed action before execution.
Okta, Entra, Auth0, StrongDM and Delinea control who can act and what they can reach.
They don’t decide whether this exact move is allowed.
They confirm who is acting; Okren decides whether this move is allowed.
Cerbos, Oso and Permit.io write down the rules and who controls them.
They don’t check each live move against the full situation.
They store the rules; Okren applies them to each move as it happens.
Microsoft and Faramesh protect systems from attack and misuse.
A secure system can still take a move nobody authorised.
They keep attackers out; Okren checks the moves that come from inside.
Sardine, Stripe Radar and Forter judge how risky or suspicious a move looks.
A low-risk move can still be outside your authority.
They score the risk; Okren decides whether the move is allowed.
ChatGPT, Claude and Grok focus on how the AI plans and acts.
They don’t independently approve what the AI helps produce.
They help the AI act; Okren independently approves the move before it happens.
Datadog, Splunk, New Relic, Vanta, Drata and AuditBoard show what is happening and what happened.
They don’t stop the wrong move before it becomes real.
They report what happened; Okren decides before it happens.
Stripe, bank APIs, exchanges, Fireblocks and Increase make the money move.
They don’t know whether your institution should make the move.
They carry out the move; Okren decides whether it should happen at all.
Is this exact agent move allowed, right now, given everything happening across the institution, under the authority it has set?
Okren is built to independently supervise AI agents, so humans are always in control of AI moving money.
One dashboard, in your own environment.
The demo above is the dashboard. Pick an area and it jumps to it.
A supervisor does three things. Okren helps with each.
Understand
Know what is being asked
Knows who is asking, what for, how much, and where the money is going.
Brings the context of each proposed move together.
Apply authority
Check it against the limits
Checks the move against the limits the institution set, and signs off or sends it back.
Applies those limits the same way every time, at AI speed. Exceptions go to the person who owns them.
Remember
Keep the record
Writes down what was approved and why, so it can be explained later.
Keeps that record automatically.
Give it access. It builds the map on its own.
Okren starts with one money move and read access to your logs and systems. Nothing else. It follows every connection outward and learns what each one touches.
ReadPayment requestAI agent$18,400
FoundRecipientSupplier
LinkedInvoiceBudgetPayment historyYour limits
MarkedWhere money leaves
LinkedOther agents that can pay the same supplier
MappedExisting controlsTheir logs
Learning the web gives Okren no permission to act. You set the limits.
Okren can understand the whole web. It only decides the move. Illustration using an example payment.
Between whoever proposes and the system that executes.
Okren runs as software inside the institution’s own environment, in front of its execution system.
Your model, your agent framework and your execution system can change without rebuilding your rules.
- Started by
- PAY-AGENT-01 (AI agent)
- Type of move
- Supplier payment
- Amount
- $18,400
- Recipient
- Harbor Supply Co
- Answer
- Reduce to $10,000
- Why
- Over the $10,000 payment limit
- Record
- Decision ID, time, the limit that applied
Your system executes $10,000
The call
Before it executes, your system sends Okren the proposal: action, amount, recipient, account, intent, who or what is asking, and a trace ID.
The answer
Okren checks it against fixed rules, not a model’s guess, so the same input gets the same answer. It returns Approve, Reduce (with the permitted amount), Ask or Stop, the limit that applied, and a decision ID.
Held, not skipped
Your system executes only what comes back allowed. If Okren cannot answer, the move is held, not executed.
Your existing controls stay where they are
- Sanctions screening
- Authentication
- Balance checks
- Dual control and callbacks
- Reporting duties
Okren does not judge what is suspicious.
The institution sets the limits. Okren applies them consistently.
- What may move. What each person, system or AI is allowed to move, and to whom.
- How much. The size of one move, and the total for the day.
- When a person steps in. The amounts and hours that send a move to someone.
Okren applies these the same way every time, and each answer carries a 0 to 100 limit score so people know what to look at first.
Example only. Not a recommended policy.
Know what was done with money, and what was allowed.
Built to answer the question a reviewer asks later: why was this allowed to leave?
Nothing proposing a move is the one writing its own history. The record also shows who reviewed each exception, how long they took and what they changed.
Phone view shows one decision. Open this page on a desktop to see the full record.
Example decisions with sample data.
Built so the control can’t be talked out of its job.
What your security and risk teams will ask. We would rather tell you first.
In place today
What Okren does now.
- Runs as software inside the institution’s own environment
- Independent of the system making the move: it cannot approve its own answer or define its own authority
- Approve, Reduce, Ask or Stop against your limits
- Every answer recorded and reconstructable
- Identifiers can be masked in the record and in exported logs
- Held, not executed, if Okren cannot answer
- Typical check under 1 ms (current environment)
Customized for the institution
What is specific to your organization, agreed before Enforce.
- Approvals started by your people, and who approves them
- Which of your systems and payment routes pass through Okren
- Who is alerted when a payment is held, and how it is released at a cutoff
- How a change to a limit is approved
- What data enters Okren, which fields are masked, and where it is kept
Not yet
Stated plainly.
- SOC 2
- A full vendor-risk package
Okren supervises what is routed through it. A path that skips it is not supervised, so the first step is mapping where your moves execute.
Okren earns its authority in three stages.
Nothing is switched on blind. You see what Okren would do before its answers are binding.
Discover
Map where money can move, and who starts each move. Okren watches. Nothing about execution changes.
Shadow
Okren records what it would have answered. Your existing process still runs the move. You compare the two.
Enforce
Okren’s answer is binding. Moves inside the limits proceed. Exceptions are stopped, reduced or held.
Shadow on one path is the easiest place to begin.
Run one workflow in Shadow for a fixed period, compare Okren’s answers with what actually happened, and decide from the evidence.
Moving from Shadow to Enforce is a risk decision, made by whoever owns your operational risk limits. It is not a settings change.
Two fair
questions.
Why can’t an institution build it themselves?
A check built inside the team it checks is never fully independent.
Writing the rules is the easy part; the hard part is a record an auditor trusts, across every system, kept up to date.
Why not rely on the limits the institution or its providers already set?
Each provider’s limits only work inside that provider.
Okren holds one set of limits across all of them, separate from the AI that proposes the move.
People move money. AI moves money. Okren supervises BOTH.
Show us one release path that already moves money. We’ll map where an independent check would sit relative to your authorization matrix and dual-control points.