AgentPay Bridgev2026-04-17

Track 01 - AI Growth and Agentic Commerce | Payment rail: Razorpay Orders API

Landing Page
Backend Rail Live
Payment Rail: Razorpay
1-Click Protocol Test Suite // Zero Terminal Required

Trigger real autonomous AI buyer agent transactions against the adapter to verify guardrails and live Razorpay Orders.

Judge & Evaluator Testing Guide

60-Sec Read

How to test this live demo with 1-click or from the terminal.

What this project does: AgentPay Bridge is a checkout safety adapter for autonomous AI buyer agents. It translates agent purchasing intents into verified orders on the Razorpay rail while mathematically enforcing server catalog prices, 50% discount caps, 30-min inventory holds, and tamper-proof audit trails.

OPTION A: 1-Click Frontend Testing (Fastest)

You don't need a terminal! Use the buttons in the action bar above:

  • ▶ Happy Path: Creates a session and creates a live Razorpay test order.
  • 🛡 Test Attack: Dispatches rogue agent with 75% discount; stops the breach.
  • ⚡ Idempotency: Replays identical token; returns cached session without double charges.
OPTION B: Terminal Simulator Command

Run the headless buyer agent directly against the live Cloud Run backend:

python buyer_agent_sim.py --base-url https://razorpay-acp-adapter-922729192321.asia-south1.run.app

What Each Simulator Act Demonstrates & What to Look For

ACT 1: Discovery

Agent resolves merchant capabilities at /.well-known/agent.json and retrieves the 5-SKU catalog.

ACT 2: Happy Path

Agent buys items, reserves stock, and settles on Razorpay.

✦ Look for: New green “completed” row with a real Razorpay Order ID.
ACT 3: Attack Suite

Malicious agent attempts a ₹1 price override and 75% discount.

⚠️ Expected: Shows “rejected”. This is correct defense behavior, not a bug!
ACT 4: Resilience

Tests duplicate request idempotency, cancellation, and Razorpay refund.

✦ Look for: “cancelled” and “refunded” terminal state rows.
No simulator run? The dashboard displays real-time metrics, recent state transitions, and audit records automatically. Click any session row in the table below to inspect its itemized GST math and immutable Firestore audit trail.
Total Sessions
0

Lifecycle executions recorded

Completed
0

Razorpay orders created

Bounded Rejections
0

Gated violations stopped

Guardrail Bounds
Active and Enforced

Deterministic rule engine

Checkout Sessions

Live agentic checkout sessions across lifecycle

No checkout sessions found.

Run python buyer_agent_sim.py to execute the buyer agent simulation.

ACP Guardrail Rules

Deterministic bounds checked before state transition

Max Discount Allowed50%
Max Single Order ValueRs 50,000
Max Quantity / Line Item10 units
Idempotency-Key HeaderEnforced and Cached
Capability Feed (/.well-known/agent.json)

Live Audit Feed

Recent Firestore state transitions

/audit_entries

No audit entries logged yet.