Every order is a loop that runs to its outcome.
One direct to consumer order, as the engine ran it
Waiting on pickup.collectedCheckout
checkout#1
Web
Validate
validate#1
1 line · Ada Lovelace
List price
price#1
Subtotal $349.00
Tax
tax#1
Tax $27.92 · total $376.92
Authorize payment
authorize#1
Authorized $376.92
- ok (taken)
- declined
Route fulfillment
route#1
Pick up at SoHo store
- ship
- pickup (taken)
- carry_out
- digital
- unavailable
Hold at store
reserve_pickup#1
Reserved 1 line
- ok (taken)
- backorder
Ready for pickup
ready#1
Pickup code QGFZU3
Pickup ready
ready_note#1
→ ada@example.com: Web order #10655 (pickup) is ready at SoHo store
Customer collects
collect#1
Waiting for pickup.collected
- received
- timeout
- Subtotal
- $349.00
- Tax
- $27.92
- Total
- $376.92
tpl-omnichannel-order run by the engine in workerd on the dashboard’s “Buy online, pick up in store” scenario: a direct-to-consumer web order held at the SoHo store, waiting for the customer to collect it
. 8 other nodes were skipped: the branches this order didn't take.One B2B order, as the engine ran it
Waiting on payment.receivedPO received
po_in#1
Document po.txt
Extract order
extract#1
3 lines · Northwind Traders (heuristic)
Validate
validate#1
3 lines · Northwind Traders
Contract pricing
price#1
Subtotal $3,286.00 · 3 contract
Tax
tax#1
Tax $271.10 · total $3,557.10
Credit check
credit#1
Within credit ($88,000.00 available)
- ok (taken)
- over_limit
Assess: routine?
assess#1
routine 0.50 · ambiguous 0.50 · urgent 0.50 (heuristic)
- yes
- no
- unsure (taken)
Agent: triage
router#1
→ auto_approve
- auto_approve (taken)
- needs_approval
- ask_customer
Reserve stock
reserve#1
Reserved 3 lines
- ok (taken)
- backorder
Invoice
invoice#1
net_30 · due 2026-11-12 · on account
Ship
ship#1
FREIGHT FREIGHT1188453459
Post to ERP
erp#1
NETSUITE-lgmyhwwo
Payment received
paid#1
Waiting for payment.received
- received
- timeout
- Subtotal
- $3,286.00
- Tax
- $271.10
- Total
- $3,557.10
tpl-b2b-order run by the engine in workerd on the dashboard’s “B2B PO by email” scenario, recorded without a model (hence “heuristic” on
the extract step, which found cus_northwind from the email’s
From: line)
What it runs
Section titled “What it runs”Every commerce model: B2B, B2C, direct to consumer, marketplace, subscription, procurement and replenishment, from any channel. Each model starts from a built-in flow you can open in the editor, change and republish.
- Flows are data. A flow is a JSON graph of catalog nodes. Publishing makes an immutable version, and each order pins the version it started on.
- Orders are durable. Each node runs once and its result is kept, so an order survives restarts, waits days for a person or a payment, and rolls back its reservations, holds and shipments when it fails.
- Every order is live. The order page shows each node as it runs, and approvals and events reach the waiting node from the dashboard, the API, a signed webhook or the assistant.
- AI never blocks an order. Every model call has a schema, a 20-second cap and a deterministic fallback.