Agent Rollback Designer — Guide

When an AI agent initiates a payment, it executes a chain of steps across multiple systems — charge the card, create the order, update the ledger, send the receipt. If any step fails midway through, you need a reliable way to undo the completed steps and leave the system in a consistent state. This is the Saga pattern.

This tool lets you drag-and-drop the steps of your payment flow, assigns a compensating action to each step, and generates a complete AWS Step Functions state machine definition (JSON) with retry policies and compensation chains that you can deploy directly.

1. Enter a Saga Name in the top field. This becomes the Step Functions state machine name and the Lambda function name prefix in the generated JSON.
2. Use lowercase hyphen-separated names: checkout-payment-saga, refund-saga.

1. Click any step template from the left panel to add it to your flow. Steps appear in the order you add them.
2. Each step card shows its forward action and its compensating action. The compensation chain is automatic — you do not configure it separately.
3. Drag steps to reorder them. The compensation logic updates automatically to reflect the new order.
4. Click the × on any step to remove it from the flow.

1. Choose Aggressive, Balanced, or Conservative retry policy. This applies to all forward steps in the generated state machine.
2. Compensation steps always use aggressive retries regardless of this setting — a failed compensation is a critical failure that must be retried hard.

1. Click Export JSON to download the Step Functions state machine definition.
2. In the AWS Console: go to Step Functions → State Machines → Create → paste the JSON in the definition editor.
3. Via Terraform: use aws_sfn_state_machine with definition = file("your-saga.json").
4. Replace placeholder Lambda ARNs with your actual function ARNs before deploying.