Some links here may earn us a small commission. It has no bearing on what we recommend, and we have turned down partnerships that would have.

A 2026 buyer's guide to Swap, Rye, and the native checkout flows built into ChatGPT and Perplexity. Pricing verified September 10, 2026.

Treat what follows as the working session we run with a client the first time this question lands on the roadmap, minus the calendar invite. It should take about seven minutes, and by the end you will know which layer of this stack to build first, what Swap, Rye, and the native protocol inside ChatGPT and Perplexity each cost, and what a product feed has to look like before an assistant will trust it with a purchase.

What this guide covers

• Why the order-volume threshold that governs most vendor decisions bends hard for a high-ticket brand, and what that number is really a stand-in for.

• The feed and checkout requirements ChatGPT and Perplexity publish today, and what they mean for a Shopify catalog specifically.

• Swap, Rye, and the native assistant checkout in full, with every dollar figure each one puts in writing and a plain note where none exists.

• A side-by-side table, then a recommendation sorted by the brand you actually run rather than the brand a sales deck imagines.

• The questions founders ask us about this, answered directly, including whether a return can be handled by the same assistant that made the sale.

The short version. A brand well under a thousand orders a month with a clean, accurate Shopify feed already qualifies for the checkout flow ChatGPT and Perplexity run at no charge once approved, and that free layer belongs in the build plan before anything else, since it is the floor every paid option sits on top of. Rye suits a team that would rather meter agentic checkout the way it meters any other API, a flat monthly base plus a few cents a transaction, instead of negotiating a contract for something this new. Swap suits a brand that wants to buy the finished conversational storefront outright, cross-border pricing, returns, and the assistant layer bundled into one system, earning its price once running duties, carriers, returns, and a chatbot as four separate vendors costs more in seams and staff time than the platform itself would. None of the three replaces the groundwork, since an assistant can only sell what a brand's own data lets it see clearly.

Readiness is the real gate, not revenue

The volume figure worth anchoring on is roughly a thousand orders a month, the point where Rye's per-order fee starts to rival the flat cost of a fuller platform and Swap's cross-border layer starts earning back its price in duty and return handling alone. Read that number loosely. A candle brand shipping four hundred orders a month at an average order value of eleven dollars sits nowhere near that line, but a bridal or fine jewelry brand moving sixty orders a month at four thousand dollars apiece already carries more transaction risk than its order count suggests, and should be sized by dollars moved rather than orders placed. Mid-market, for this guide, runs from two million to fifty million dollars in yearly revenue, and a thousand monthly orders is simply where several vendors' tiers happen to reprice, not a number with physics behind it.

Before that arithmetic matters, the feed has to hold up. An assistant reads a structured file describing products, prices, availability, and policy rather than browsing a storefront the way a person does, and will not transact against anything it cannot parse with confidence. A brand missing identifiers or running stale inventory counts is effectively invisible here regardless of what it later spends on Swap, Rye, or anything else, which makes a feed audit the real first purchase in this category.

What the protocol actually requires, and what the schema already assumes

OpenAI publishes its Agentic Commerce Protocol and checkout specification openly at developers.openai.com, worth reading before signing anything with a vendor promising to handle this. A merchant that wants ChatGPT's Instant Checkout to complete a purchase stands up three REST endpoints, one to open a checkout session, one to update it as a shopper changes an item or a shipping method, and one to complete the order, alongside a webhook reporting order status so the assistant never tells a shopper something the warehouse cannot back up. Payment still runs through the merchant's own processor; the protocol governs the conversation about the cart rather than the movement of money. The catalog side calls for nine required product fields at minimum, in a Google-compatible feed format, before an item is eligible to be discovered in a chat, let alone bought there. Instant Checkout remains invitation-only for approved partners, and OpenAI's documentation lists no merchant fee for participating.

None of that structure is new so much as inherited. Schema.org has defined the Product and Offer vocabulary machine readers of nearly every kind lean on for well over a decade, and Google's own structured data documentation for product markup, the reference deciding whether a search listing earns a price and a star rating, is functionally the training ground for what an assistant now expects a product page to say about itself. A brand that already did this work for search visibility is most of the way to agentic readiness without spending a dollar on any option below; a brand that has not is buying software to paper over a gap clean structured data would have closed for free. Perplexity runs a comparable buying flow today, extended to a smaller set of merchants, and this guide treats it alongside ChatGPT's flow rather than pricing it separately, since Perplexity has not published documentation with the same level of detail.

The three layers of the stack, and where each one sits

Who actually owns the checkout session. Swap and the native flows both keep the transaction inside the assistant's own interface from question to receipt, while Rye's API can be wired either way, which matters to a brand that wants the payment screen to stay on its own domain.

How the cost is measured. The native protocol charges nothing beyond engineering time, the only fixed-cost option regardless of volume. Rye prices like infrastructure, a flat base plus cents per order and per lookup, growing with consumption. Swap prices like a platform, a fee negotiated to a brand's footprint, trading a metered relationship for a fixed one.

What happens once a border enters the picture. Landed cost, duty calculation, and multi-currency pricing sit natively inside Swap because cross-border operations were its starting point before the conversational layer was added; Rye and the native protocol both assume a merchant already handles tax and duty through its own stack.

Whether the assistant can finish more than the sale. A return or a store-credit request is still commerce, and only Swap folds that after-sale conversation into the same assistant relationship that handled the purchase.

Swap: the conversational storefront built to finish the sale

Verification attempted September 10, 2026 against swapcommerce.com; both the marketing site and its pricing path returned a blocked response rather than a public rate card, consistent with how the platform has historically sold enterprise cross-border deals through a sales conversation rather than a self-serve signup.

Swap suits a brand that would rather buy one coherent system than assemble four separate contracts for duties, returns, shipping, and an AI storefront layer and wire them together itself. Its pitch is consolidation: a shopper who asks an assistant to buy something, gets quoted the right price including duty, and later comes back to exchange it, stays inside one system the entire time instead of bouncing between a checkout vendor, a duty calculator, and a returns portal sharing no customer record. The tradeoff is that a brand cannot try Swap with a credit card and a trial the way it can try Rye; the relationship starts with a quote, rewarding a brand that already knows its own volume and market mix.

[Partner quote pending, requested via Abel, due September 30, 2026.]

Name, Title, Swap

Rye: the metered engine behind someone else's assistant

Verified live at rye.com/pricing on September 10, 2026.

Rye treats agentic checkout as an API a development team can meter and monitor the way it meters any other paid service, which is why technically staffed brands reach for it first. A hundred and forty-nine dollars a month buys fifty dollars of usage before per-order and per-fetch charges accrue, and pricing scales with consumption rather than revenue, so a small brand doing low hundreds of agentic orders a month can run this for well under two hundred dollars total. What it leaves out is the operational layer: duty calculation, a returns portal, and a bundled storefront stay with the brand's existing stack, a tradeoff a developer-led team usually welcomes for full control of checkout logic.

The native flow: what ChatGPT and Perplexity already do for free

Verified live at developers.openai.com/commerce (overview, checkout specification, and get-started guide) on September 10, 2026. No third-party pricing was used; figures reflect OpenAI's own developer documentation as read today.

This is the option most brands do not realize they already qualify to build, since it costs nothing beyond engineering hours and the discipline to keep the feed current. The catch is availability rather than price: Instant Checkout stays gated to approved partners, and the application and review process is the real bottleneck, not a subscription fee. For a brand under the volume threshold above, this native layer alone may be the whole investment worth making this year, and it is worth building even if Rye or Swap gets added later, since the feed work it forces is the same work every paid layer eventually requires.

Side by side, on one page

Pricing verified September 10, 2026 from each vendor's own published pricing page or developer documentation, as cited above.

Which one to build first, by the brand you actually run

A lean DTC brand still finding its footing, doing a few hundred orders a month with no dedicated engineering headcount, gets the most out of doing the feed work the native protocol requires and applying for Instant Checkout, since that single move costs nothing beyond staff time and forces the catalog discipline that matters for search visibility regardless of eventual volume. A mid-market Shopify Plus brand with a real development team, inside the two-million-to-fifty-million-dollar band and past a thousand orders a month, is the clearest fit for Rye, since metered pricing keeps pace with a channel still proving itself. A brand running volume across multiple countries, or a lower-volume, high-average-order-value brand for whom one mishandled cross-border return can cost more than a month of subscription fees, should be having the Swap conversation now, since the value sits in consolidating duties, returns, and the assistant layer under one roof.

Questions we get asked

What is agentic commerce, exactly?

Agentic commerce is a purchase completed by an AI assistant acting on a shopper's behalf inside the assistant's own interface, from discovery through payment, rather than the assistant recommending a product and sending the shopper elsewhere to buy it.

Can ChatGPT actually complete a purchase today, or does it just recommend products?

For merchants approved into Instant Checkout, ChatGPT can complete the entire purchase inside the chat using the Agentic Commerce Protocol's checkout endpoints, though most brands outside that group still only receive recommendations that route a shopper back to their own site.

What does a brand have to change on its product feed for an AI assistant to sell it?

At minimum the feed needs nine required product fields in a Google-compatible format covering identity, pricing, availability, and variants, since an assistant will not recommend or transact against a product it cannot parse with confidence.

Is Swap only useful for brands selling internationally?

No, but its pricing and its value are both built around cross-border volume, so a purely domestic brand will typically get more return per dollar from Rye or the native protocol first.

How much does it cost to let an assistant buy on a customer's behalf?

It ranges from no published merchant fee for the native flows to a metered API starting at $149 a month with Rye to a custom quote with Swap, and the right choice depends more on order volume and market complexity than on any single price.

Does letting an assistant complete a sale change who owns the customer relationship?

The order and customer record still belong to the merchant under all three approaches here, since payment runs through the merchant's own processor and the assistant completes a transaction on the merchant's data rather than reselling on its own account.

Can a customer return something an AI assistant sold them?

Yes, though only Swap currently handles that return inside the same assistant relationship that made the sale; under Rye or the native protocol, the return follows the brand's ordinary process just as it would for any other order.

How does this connect to the answer engine optimization work you already do for us?

The same structured product data and clear policy language that make a catalog legible to an answer engine are the exact inputs an assistant needs before it will complete a purchase, which is why our AEO work and this readiness work now sit on the same checklist.

Where to go next

If your catalog has never been audited against a structured data standard, that is the honest starting point regardless of which of these three you choose, overlapping heavily with the groundwork behind our SEO and AEO engagements. If your team is weighing a Shopify Plus replatform or a checkout rebuild in the next year, this is the moment to design the agentic layer in rather than bolt it on afterward, exactly the kind of question our CRO and experience team fields regularly alongside our work with Swap and Shopify. Get in touch and we will walk your feed, your volume, and your market mix against this framework before you commit to any of the three.