Policy
The rules the gate enforces
Editable. Every change rebuilds the policy set that the engine evaluates against. Amounts are in whole US dollars for readability; the engine works in integer minor units under the hood.
Interactive · No install
Watch the gate decide, live.
Propose a payment and send it through the same deterministic policy engine that powers Axiru. It runs entirely in your browser, on sample data, and answers the one question that matters: may this agent move this money, right now, at this amount. Allow, require approval, or deny, every time with a stable reason code and a decision fingerprint you can replay.
This runs the real deterministic engine in your browser, on sample data. No money moves, no account, nothing is sent anywhere. The same evaluator, reason codes, and SHA-256 decision fingerprints power the hosted product. The hosted product adds what a browser cannot: credential custody, human approvals across live rails, and a persisted hash-chained ledger. The session ledger below is in memory only and clears on reload.
Propose a payment
What the agent wants to do
Fill in a spend intent and send it through the gate.
Verdict
No decision yet
Send a proposal or pick a scenario. The engine returns allow, require approval, or deny, with a stable reason code and a decision fingerprint.
Preset scenarios
The story in four clicks
Session decision ledger
Every decision, with its fingerprint
A demonstration of the tamper-evident-record idea. In the hosted product this is a persisted, hash-chained ledger. Here it lives in memory for this session only.
Nothing yet.
What you are looking at
The engine reads the numbers, not the prompt.
A spend control is not a filter on what the agent is told. The instructions are exactly what an attacker manipulates. This gate is a deterministic policy decision the agent cannot talk its way past: it evaluates the amount, the counterparty, the agent, and the time, and it returns the same verdict every time for the same inputs.
The playground uses the real evaluator from the open source packages. The hosted product continues where a browser cannot: it holds credentials, routes approvals to named people across live rails, and writes every decision into a persisted, hash-chained evidence ledger. See the open source page for the split between what runs in your process for free and what runs on our infrastructure.
Next step
When you want the parts that run on infrastructure, that is the hosted product.
Credential custody, a real approval queue with named approvers, multi-rail enforcement, and a hash-chained evidence ledger. Axiru Agent Controls is the hosted continuation of the engine you just used.
Start in shadow mode first. Move to live enforcement later.