WooCommerce order-status chat agent with DHL tracking

WooCommerce order-status chat agent with DHL tracking

An n8n workflow that puts an AI agent behind a chat on a WooCommerce shop. A customer asks about an order. The agent looks the
customer up by the email address your website supplies, reads their orders through the WooCommerce API and adds DHL tracking
events where an order has a tracking number. The address reaches the workflow encrypted, so the chat answers only about the
customer it was opened for.

What is inside

• workflow.json: the workflow as n8n exports it, forty nodes. Thirty-one of them run and nine are sticky notes. It holds no
  credential.
• README.md: these notes.

What we changed from the original copy

• OpenAI Chat Model called gpt-4, which OpenAI shut down on 23 October 2026. OpenAI's deprecations page, read on 12 November
  2026, names gpt-5.6-sol as its substitute, so the node calls that now.
• The agent's system message told it the customer's address with {{ $json.email }}. The chat item has no such field, because
  Mock Data and the decrypt step both put the address under metadata.email. The message reads {{ $json.metadata.email }}
  now, so the agent can repeat the address when asked.
• WooCommerce Get Orders pointed at the author's own test shop. Its address is https://YOUR_STORE_DOMAIN/wp-json/wc/v3/orders
  now.
• The three Code nodes that encrypt and decrypt an email address used the password a random password. Each now holds
  YOUR_SHARED_PASSWORD.
• The sample customer address, and its encrypted form, are placeholders in Mock Data, Encrypt email, Example encrypted email
  and the example page: <CUSTOMER_EMAIL> and <ENCRYPTED_EMAIL>. The page's chat address is <YOUR_CHAT_WEBHOOK_URL>.
• The file had no name, so we gave it one. The webhook IDs and the credential links are gone.
• Nothing else. The Code nodes' logic is as the file had it.

Before you start

• An n8n instance, on n8n Cloud or your own server. The Code nodes use Node's crypto module. The file's own note says the
  environment variable NODE_FUNCTION_ALLOW_BUILTIN must equal or include crypto, and calls that the default on n8n Cloud.
  Check it on a server of your own.
• A WooCommerce shop and a WooCommerce credential for its REST API, with a consumer key and secret.
• An OpenAI API key.
• A DHL credential, if you keep the tracking branch.
• Order tracking numbers stored in each order's meta data under the key _wc_shipment_tracking_items. The tracking branch reads
  that key.
• A website where the chat will sit, and a backend that can encrypt the customer's email address.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Select your WooCommerce credential in WooCommerce - Get User and WooCommerce Get Orders, your OpenAI credential in `OpenAI
   Chat Model, and your DHL credential in DHL`.
3. In WooCommerce Get Orders, replace YOUR_STORE_DOMAIN with your shop's domain.
4. To test, open Mock Data and replace <CUSTOMER_EMAIL> with the email of a customer who has orders. As the file ships,
   Mock Data is on and Decrypt email address is off. Press the Chat button and ask about an order.
5. Open AI Agent and edit the system message. It names a shop, Best Shirts Ltd., and states its dispatch and delivery times.
6. For production, choose one password and put it in place of YOUR_SHARED_PASSWORD in Decrypt email address. In your own
   backend, encrypt each signed-in customer's address the way Encrypt email does, with the same password, and give the result
   to the chat as metadata.email.
7. Enable Decrypt email address and disable Mock Data. Disable it and do not delete it: the key in Window Buffer Memory reads
   the session ID from it. If you delete it anyway, point that key at Chat Trigger.
8. Publish the workflow. Copy the production address from Chat Trigger into your page where it says <YOUR_CHAT_WEBHOOK_URL>.
   Webhook Example Page, with the path website-chat-example, serves a sample page you can copy.

How a run goes

A chat message:

1. Chat Trigger is public, so anyone with its address can open a chat. It passes on the message, the session ID and the
   metadata object the page set.
2. Decrypt email address turns metadata.email back into an address. It is off in the file. Mock Data sets that field to a
   fixed address instead, and it is on.
3. AI Agent (version 1.4) runs with OpenAI Chat Model, Window Buffer Memory, which keeps the last 10 messages of a session,
   and one tool, WooCommerce_Tool. Its system message sets the tone, says to use only the customer's own address, and states
   that orders are dispatched 1 to 2 days after purchase and delivered 1 to 2 days later.
4. WooCommerce_Tool calls this same workflow with the field email, taken from metadata.email. The agent passes the order IDs
   it wants, comma separated, or nothing for every order.

The tool call, which starts at Execute Workflow Trigger:

1. If email provided stops an empty address at No email provided.
2. WooCommerce - Get User finds one customer with that email. If user found sends a miss to No customer found.
3. WooCommerce Get Orders asks /wp-json/wc/v3/orders for that customer's orders, limited to the agent's IDs when it gave any.
   If order found sends a miss to No order found.
4. Extract Tracking Data reads _wc_shipment_tracking_items from each order. If contains DHL data passes only orders with at
   least one tracking entry, and Split Out makes one item for each.
5. DHL looks up each tracking number. Its error output goes to Add Error Information, which writes that no data could be found
   for the parcel. Merge Tracking Data and Aggregate gather the results into tracking.
6. Merge Orders and Merge Order and Tracking Data join each order with its tracking, and Send Response hands the agent one
   response list. A miss at any step above puts its short sentence in response instead.

Change the look

• Shop name, tone and delivery times: the system message in AI Agent.
• The model: OpenAI Chat Model. gpt-5.6-terra is the cheaper of OpenAI's two named replacements.
• How much the chat remembers: Context Window Length in Window Buffer Memory.
• Another carrier: replace DHL with that carrier's lookup, or delete the tracking branch and keep only the orders.
• The page: the HTML in Respond to Webhook.

Good to know

• Only the encrypted value ties a chat to a customer. It has no signature and no expiry, and whoever learns the password can write
  a value for any customer. Keep the password out of the page and out of the repository.
• Chat Trigger is public, so anyone who finds its address can start a chat on your OpenAI key.
• Every tracking number goes to DHL, whatever the carrier. A parcel from another carrier ends at Add Error Information.
• With Mock Data on, every chat belongs to the one address typed there. Turn it off before the chat meets real customers.
• The example page loads the chat widget from jsDelivr as @latest. Pin a version on your own site.
• Where the data goes: each lookup sends the whole order to OpenAI, with the customer's name, addresses, phone number and items, and
  the tracking numbers go to DHL.
• Billing: OpenAI charges for each agent turn, and each lookup adds a call to your shop's API and one to DHL for each parcel.