Live demo — deterministic sequence enforcement for an AI agent

Press a button below and an AI agent runs that scenario against the live gate. Nothing runs until you start it. The gate decides whether each step is permitted before execution — not the agent itself. Green means the step ran in correct compliance order. Red means the gate blocked it: a duplicate request, a skipped approval, or an attempt to reopen a closed file. Every decision produces a cryptographic receipt written before any action executes, and every run ends with a one-click link to its signed report — verifiable without relying on the agent's own logs.

This is the public demo lane. Anyone can read the report for a demo- sequence without a key, so whatever a run puts in attestation is public. Demo receipts are also deleted 30 days after they are written. Both are fine for a demo and wrong for real work — get your own key for a private lane whose receipts are kept.

ISOLATION ORDER REPLAY SEAL BINDING PAYLOAD DEPLOYMENT

SCENARIOS

⤼

Skip to Money Movement

The identity check passes. The agent then jumps straight to moving funds — skipping the risk assessment and manager approval entirely. The gate blocks it. Money doesn't move without the checks.

EXPECT: DENY ORDER LAW ✗

verify_id → assess_risk → request_approval → execute_transfer ✗

⟳

Replay Attack

The identity check passes. Then the agent sends the exact same request again — identical nonce, identical payload. The gate recognises the duplicate and blocks it immediately.

EXPECT: DENY REPLAY LAW ✗

verify_id → verify_id ↩

🔒

Sealed Sequence Breach

A loan runs all 8 steps to completion and seals. Then the agent tries to re-enter the closed sequence. Once sealed, nothing can reopen it — not even the same agent that ran it.

EXPECT: DENY SEAL LAW ✗

verify_id → ··· → settle_loan → BREACH ✗

✓

Complete Loan Approval

A $25,000 loan runs every required step — identity check, risk assessment, manager approval, fund transfer, compliance verification, audit ledger entry, party notifications, and settlement. Real attestation evidence is attached at each step and cryptographically signed into the receipt chain. Seals when complete.

EXPECT: ALLOW ATTESTATION

verify_id → assess_risk → request_approval → execute_transfer → verify_compliance → audit_ledger → notify_parties → settle_loan

EXPLORE — SEND A CUSTOM REQUEST

⚡

CUSTOM REQUEST

Send any payload you want against the live gate.
Try replay attacks, wrong step order, fake sequence IDs, anything.

LIVE GATE YOU CONTROL

SEQUENCE_ID

STEP

MODEL_ID

NONCE (auto-rotates after each send)

ACTION (override)

ACTION_TYPE (override)

TRY:

OUTGOING PAYLOAD

GATE RESPONSE

LIVE GATE LOG

No requests sent yet.
Run a scenario to see live gate responses.

0 This session

0 Allowed

0 Denied

— Last ms

ISOLATION · ORDER · REPLAY · SEAL · BINDING · PAYLOAD · DEPLOYMENT

You just ran that. Here it is in your code.

The same gate, the same public demo key, no account and nothing to sign up for. You declare the order once; the gate refuses anything out of it and signs a receipt for every decision either way.

pip install agenticrail

# No key needed. With no argument the client sends no credential and the
# gate answers on its public demo lane, saying so in every decision.
from agenticrail import RailClient

client = RailClient()

seq = client.sequence(
    sequence_id="my-agent-run-001",
    step_order=["verify_identity", "assess_risk", "execute_transfer", "audit_ledger"],
)

seq.next("verify_identity")    # -> ALLOW
seq.next("assess_risk")        # -> ALLOW
seq.next("execute_transfer")   # -> ALLOW
seq.next("audit_ledger")       # -> ALLOW, seals the sequence

# Out of order, replayed, or after the seal: raises RailDenied
seq.next("verify_identity")    # -> RailDenied: SEALED_SEQUENCE

npm install @agenticrail/core

// No key needed. With no options the client sends no credential and the
// gate answers on its public demo lane, saying so in every decision.
import { RailClient } from "@agenticrail/core";

const client = new RailClient();

const seq = client.sequence("my-agent-run-001", [
  "verify_identity",
  "assess_risk",
  "execute_transfer",
  "audit_ledger",
]);

await seq.next("verify_identity");   // -> ALLOW
await seq.next("assess_risk");       // -> ALLOW
await seq.next("execute_transfer");  // -> ALLOW
await seq.next("audit_ledger");      // -> ALLOW, seals the sequence

// Out of order, replayed, or after the seal: throws RailDenied
await seq.next("verify_identity");   // -> RailDenied: SEALED_SEQUENCE

Works under LangGraph, CrewAI, Mastra, Genkit or a plain loop — it is one HTTPS call before each step, so the framework is irrelevant. Full reference in the docs.

WHAKAIRO AS LAW

Built from whakairo — Māori carving philosophy. The chisel has no ctrl+z. Every committed action has consequence. The gate applies that discipline to AI agents. Sequence is law.

HOKIANGA · AOTEAROA