Automated purchasing

“Order sushi from my usual place for under $40.”

A purchase session turns a sentence into a priced cart, a policy decision, an approval when needed, and a card that exists only for that order.

Try a simulated purchase See the MCP tools

Purchase session · pur_4d81

Order sushi from my usual restaurant for under $40.

Found Sakura House (your last 4 orders). Cart: 2× Salmon roll, 1× Miso.
Subtotal $28.40 · Tax $2.62 · Delivery $3.99 · Service $2.10
All-in total $37.11

Within budgetCard issuingSimulated merchant

mcp · request_purchase

{
  "tool": "request_purchase",
  "arguments": {
    "description": "Sushi dinner delivery",
    "merchant_preference": "DoorDash",
    "maximum_amount_cents": 4000,
    "delivery": { "address_id": "adr_home" },
    "reason": "Team dinner"
  }
}

// -> purchase_task pur_4d81
// -> policy: auto_approve (under $50, food_delivery)
// -> card_9fJ2 issued, merchant_lock: DoorDash

Coverage

Categories and merchant integrations

Merchant integrations are simulated in the prototype. No real orders are placed until a payment processor and merchant APIs are connected.

Purchase categories

Food delivery
Groceries
Office supplies
Software subscriptions
Domain names
Event tickets
Travel
Local services

Merchant integrations

DoorDash
Uber Eats
Instacart
Amazon
Shopify stores
Namecheap
GoDaddy
Ticketmaster

Order lifecycle

You always know where the order is

Each purchase session exposes its state, so a stalled agent is visible rather than silent.

01Gathering information
02Searching
03Building cart
04Waiting for user
05Waiting for approval
06Creating card
07Checking out
08Confirmed

Watch an agent buy something, safely

Run the full purchase flow against simulated merchants and see every policy decision it triggers.

Start in test mode View developer docs