AI agent for logistics and supply chain

How it works

1. Receive documents and updates

   Bills of lading, commercial invoices, packing lists and carrier emails arrive from your inbox, portals and EDI. Each is saved against its shipment reference.

2. Read and compare

   The agent pulls out shipper, consignee, ports, weights, pieces, commodity codes and charges, then compares the documents with each other and with the booking. A weight that differs between the invoice and the packing list is caught here.

3. Track and answer

   It follows each shipment through carrier tracking and portals, answers 'where is my order' for staff and customers, and flags delays against the promised date.

4. Update or hand over

   Clean documents update the shipment record in your TMS or ERP. Mismatches, new commodity codes and carrier invoices that differ from the rate go to an operations queue with the document image beside the extracted fields.

New to AI agents? How they work, in plain words

Real examples, setup steps and tool connections for this agent

An example conversation

An example of how a conversation runs. The grey lines show what the agent does in your systems meanwhile.

1. Customer(operations manager, Slack) What is stuck this morning?
2. Read 46 open shipments from the TMS and checked tracking with 4 carriers. Read 23 documents received overnight.
3. AgentTwo shipments need you. Container MSCU-4417 is 2 days behind: the vessel changed its port call. And booking 2290 has a mismatch: the packing list says 480 cartons, the invoice says 408.
4. CustomerCheck which one is right for 2290. The customer is waiting for the release.
5. Compared the packing list, invoice and the original booking. Booking and invoice both say 408 cartons.
6. AgentThe booking and the invoice both say 408, so the packing list is likely a typo. I have not changed anything. I can ask the shipper for a corrected packing list and tell the customer there is a short delay. Shall I?
7. CustomerYes, ask the shipper. For the container, tell the customer the new date.
8. Sent an email to the shipper asking for a corrected packing list. Sent the customer the revised arrival date for MSCU-4417.
9. AgentBoth are sent. I will check the shipper's reply this afternoon and update booking 2290 when the corrected list arrives.

How it is built

Where the input comes from

Documents come from the shared mailbox through its API, from portals and from EDI. Carrier tracking comes from carrier APIs where they exist, such as FedEx and UPS, and from a browser logged in with an account you control where a portal has no API.

Reading the documents

Scans go through OCR first, because many bills of lading are rotated or faint. The language model then fills a fixed form: shipper, consignee, ports or addresses, weights, pieces, commodity codes, incoterm and charges. Fields it cannot read with confidence are left empty and flagged.

Checks are rules

Weights and piece counts must agree across documents and with the booking. Commodity codes are checked against the current tariff table. Carrier invoice lines are compared with the contract rate. A model does not decide these. They are fixed checks that you can read.

A fixed list of actions

The agent can update a shipment record, send a status message, ask a shipper or carrier for a missing document, post a carrier invoice for approval and raise an exception. It does not rebook, change a booking or file customs entries unless you allow a defined routine case.

Customs stays with specialists

A new or changed commodity code always goes to a customs specialist. The agent prepares the data and shows its source, and a person makes the call.

Logging and review

Every extracted field and every decision is stored with the document it came from. A weekly report shows the documents handled without a touch, the exception rate and the days from receipt to posting.

Cloud or your own servers

It can run in the cloud, or on your own servers, for example on the warehouse or office network. Then the agent, its database, its logs and the language model itself stay with you, using an open model such as Llama, Qwen or Mistral, so shipping documents and rates stay inside. Reading documents already on your network needs no internet connection, while email, carrier portals and tracking still pass messages through those services. Local open models are smaller than the largest cloud models. For reading documents and extracting fields they are usually enough, and we test on your own documents first. We process only documents your company has the right to use.

What you can connect to it

TMS (Transport Management System)

Matches documents to shipments, updates milestones and charges, and reads booked rates, through the system's API or exports.

ERP (SAP, Odoo, Microsoft Dynamics 365 Business Central)

Reads sales and purchase documents and delivery orders, updates delivery status, and posts freight invoices for approval.

FedEx and UPS

Reads tracking events and shipment status through their APIs, and gets labels or tracking numbers where your account allows.

Shipping lines and freight carrier portals

Checks container and booking status. Where a portal has no API, it uses a browser with a limited account you own.

Gmail or Outlook shared mailbox

Reads incoming shipping documents, sends status updates and requests for corrected documents.

Customs and tariff tables

Checks commodity codes against the current table you use, and flags new or changed ones.

Carrier invoices

Reads invoice lines and compares each with the contracted rate before approval.

Google Sheets or Excel

Exception lists, tracking reports and rate tables for teams without a TMS.

Slack, Teams or WhatsApp

Answers status questions from staff and customers, and alerts operations about late or mismatched shipments.

Related agents

• AI agent for procurement and purchase orders
• AI agent that works in your websites and portals
• AI agent for inventory management