Agentic Commerce Protocol (ACP)
An open protocol for agent checkout. It defines the messages an agent sends to quote a cart, open a checkout session, complete the purchase with payment, and cancel — and what the store sends back at each step.
From auto spare parts to optical stores to cloud kitchens — find the configuration made for you.
Browse all industriesFBR, ZATCA, MyInvois & UAE e-invoicing — built into every invoice, not a bolt-on. Issue e-invoices from day one.
Compliance hubBook a free 30-minute demo on your own kind of business — no slides, just the product.
Book a demoAgentic commerce is shopping done by an AI agent on a person's behalf: the shopper asks an assistant such as ChatGPT, and the agent finds, compares and buys the product from the store through an open checkout protocol.
Agentic commerce means an AI agent does the shopping, not just the searching. In ordinary online shopping, a person searches, opens stores, compares and fills in a checkout. In agentic commerce the person states what they want — "a lawn suit under 5,000, delivered to Lahore by Friday" — and an AI assistant acting for them reads store catalogs, compares the options, asks the store for a price, and places the order.
For a store, that changes who the first visitor is. Before a person ever sees your product page, a machine has read your catalog, your delivery areas and your payment options, and decided whether you are worth recommending. The store has to be readable by that machine, and buyable by it.
The same four steps happen whether the agent is a general assistant or a shopping app. Each step needs something from the store.
In plain words, to an AI assistant: the product, the budget, the city, the date, the way they want to pay. The assistant turns that into a search across stores it can read.
It reads product feeds, product markup on store pages and a description of each store's policies. A store that publishes nothing machine-readable gives the agent little to go on.
Through an open checkout protocol it asks for the exact price with delivery, holds the items, passes payment or the cash-on-delivery choice, and gets a confirmation back — or cancels.
From here it is an ordinary order: confirm, pick, pack, ship, collect payment and record the sale. The agent reports the confirmation and tracking back to the shopper.
An AI agent cannot fill in a web checkout the way a person does, and every store's checkout is different. The two protocols solve that: they are open specifications that set out how an agent and a store talk through a purchase, so one agent can buy from many stores in the same way.
An open protocol for agent checkout. It defines the messages an agent sends to quote a cart, open a checkout session, complete the purchase with payment, and cancel — and what the store sends back at each step.
A second open protocol for the same job: a standard way for AI agents to discover what a store sells and to quote, pay for and confirm an order. The purpose is the same as ACP; the message shapes differ.
| Step | What the agent asks | What the store must answer |
|---|---|---|
| Discover | What do you sell, at what price, and is it in stock? | A product feed and product markup that match the real catalog. |
| Quote | What is the total for these items, delivered to this address? | The exact price, delivery cost and delivery options, at this moment. |
| Session | Hold these items while the buyer pays. | A checkout session with the stock reserved. |
| Complete | Here is the payment, or the cash-on-delivery choice. Place the order. | An order number and confirmation the agent can show the shopper. |
| Cancel | The buyer changed their mind. | The session closed and the stock released. |
A person who finds the wrong price on a website notices it before paying. An agent does not: it buys on what it was told. If your feed says six in stock and you sold the last one at the counter an hour ago, the agent still places the order — and you either cancel on a customer you never spoke to, or ship late.
Prices work the same way. A feed copied from your website once a day can be a day out of date; a promotion that ended this morning can still be quoted to an agent this afternoon. With agentic commerce, the feed is the shop window and the price tag at once, so it has to come from the system that actually holds your stock and prices.
On EloShop it does. The feed, the store description and every quote the agent checkout gives are read from EloERP — the same stock and price list your shop counter and your other sales channels use.
| Where | What happens |
|---|---|
| Shop counter | A lawn suit is sold at the Lahore branch counter |
| EloERP | Stock moves from 6 to 5 live |
| Product feed | The feed row for that product is refreshed from EloERP |
| Agent quote | Rs 4,480 · 5 in stock · delivery to Lahore by courier |
Illustrative example. Every value comes from the store's own EloERP data.
Making a store readable and buyable by AI agents means publishing a feed, a store description and product markup, and running a checkout agents can call. On EloShop every store has all of it, kept up to date from EloERP, with nothing for you to build.
One feed per store in Google Merchant format, as XML and JSON, with price, stock, variants, shipping and returns — refreshed as products, prices and stock change in EloERP.
A plain description of the store at /.well-known — policies, delivery areas, payment methods and where the feeds are — so an assistant answers "do they deliver to Multan and take COD?" correctly.
schema.org Product, Offer and AggregateRating structured data on every product page, so agents and search engines read the same facts the shopper sees.
Quote, session, complete and cancel — in the Agentic Commerce Protocol shape, with a Universal Commerce Protocol adapter over the same checkout, so price, stock and rules are identical whichever an agent speaks.
Card or wallet when an agent pays, through the store's gateways. A cash-on-delivery order from an agent is confirmed with the buyer on WhatsApp, exactly like an order from your website.
An agent's order is an ordinary EloERP sale: it reaches the order queue, moves stock, books the courier and posts to the ledger like any other order.
An assistant deciding whether to recommend your store needs answers a product feed does not carry: which cities you deliver to, whether you take cash on delivery, what your return policy is. EloShop publishes them in one machine-readable file at /.well-known/eloshop.json, next to links to the product feeds.
The file is built from the same settings your store and its own shopping assistant use, so when you change a delivery area or a return rule, what AI crawlers read changes with it.
/feed/products.xml and /feed/products.jsonWhen an agent decides to buy, it calls the store's agent checkout at /agent/checkout. EloShop answers each call from EloERP: the quote is priced from your price list and courier rules, the session holds the stock, and completing it places a real order.
Agents that use the Agentic Commerce Protocol and agents that use the Universal Commerce Protocol reach the same checkout underneath, so there is one set of prices, one stock count and one set of rules — never two versions of your store.
| Agent calls | EloShop answers from EloERP |
|---|---|
| quote | Rs 4,480 + delivery to Lahore · courier and date from your rules |
| session | Checkout session open · 1 unit reserved live |
| complete | Order #ES-10482 placed · COD · confirmation sent on WhatsApp |
| cancel | Session closed · reserved unit released to stock |
Illustrative calls on a sample store.
There is no separate inbox for agent orders and no export to re-key. The order arrives in the same queue as your website, Shopify, WooCommerce and Daraz orders and goes through the same steps: COD confirmation, courier booking, stock movement, ledger posting and courier remittance matching.
That is what makes it safe to let agents buy from you: nothing about the order is special once it is placed.
See order management| Order #ES-10482 | In EloERP |
|---|---|
| Channel | EloShop store · placed through the agent checkout |
| Payment | Cash on delivery · confirmed by the buyer on WhatsApp |
| Stock | Lahore branch · 5 left live |
| Courier | Booked from your courier rules · tracking sent to the buyer |
| Books | Sale posted to the ledger · COD remittance matched when the courier pays |
Illustrative order on a sample store.
Live feed, AI-readable store description, product markup and an agent checkout — on your real EloERP stock and prices.
No credit card needed · Pricing on request