Fraud and Risk Review

Review Every Risky OrderBefore It Ships

A rule that flags too much buries your team, and a rule that flags too little costs you money. AssistJet reads each flagged order against the customer's history and shows you the exact call to approve.

Starts From
A new order over $500
Model Use
1 agent step
Your Part
1 approval for each order

You Have to RememberWhy It Went Through

A rule flags an order. You open the customer, the payment, and the last ten orders in three tabs, then leave a note you will never find again. When an auditor asks why that order went through, you have to remember the answer.

  • Hold the order the moment it is flaggedThe automation
  • Read the customer and the paymentThe agent
  • Write what looks wrong, in plain wordsThe agent
  • Approve the exact releaseYou
  • Store every step for the next auditThe automation

One Risky Order, Held and Reviewed

When an order over $500 comes in, hold it, read the customer history, and ask me before you release it.

You Set What the Agent May Do

Each tool has one level for each seat, and your rules are enforced in code before the call. These tools are examples, and your own list comes from your app.

  • customers_historyThe agent reads past orders.Read
  • payments_getThe agent reads the payment.Read
  • orders_holdThe run holds the order.Auto
  • orders_releaseYou approve the release.Ask First

To set a tool that cannot be undone to Auto, you type its name when the page asks.

Describe It, or Ship It as a Recipe

Write the job in plain words, and the agent drafts these steps as a card. If you build apps for other people, ship the same steps as a recipe, and every customer gets the job with no setup.

assistjet_recipes
[
  {
    "id": "risk-review",
    "name": "Risk review",
    "trigger_kind": "record",
    "trigger": {
      "collection": "orders",
      "event": "insert",
      "where": {
        "field": "record.total",
        "op": "over",
        "value": 500
      }
    },
    "steps": [
      {
        "kind": "tool",
        "tool": "orders_hold",
        "arguments": {
          "id": "{{record.id}}"
        }
      },
      {
        "kind": "agent",
        "instructions": "Read this order and the customer history. List anything that looks wrong, in plain words."
      },
      {
        "kind": "ask",
        "text": "Release order {{record.id}} for {{record.total}}? {{agent}}",
        "timeout_minutes": 240,
        "else": [
          {
            "kind": "send",
            "via": "slack",
            "text": "Order {{record.id}} is still on hold. Your team declined it, or no answer came in four hours."
          }
        ]
      },
      {
        "kind": "tool",
        "tool": "orders_release",
        "arguments": {
          "id": "{{record.id}}"
        }
      }
    ]
  }
]

What You Can Count On

Approve the Exact Call

You see the exact call before anything happens. You approve it once, and it runs once.

Permissions Enforced in Code

A tool runs only when the level and the route gate both admit the caller, so a message cannot widen them.

Every Step on the Record

Each run is a thread in your inbox, and each tool call, model, and cost is stored in your database.