logo
Try for free
  • Product
  • Solutions
  • Developers
  • Resources
  • AI
  • Pricing
  • Log in
  • Create a store
  • Product

  • AI
  • Pricing
  • Try for free
  • Log In
  • Merchandising

  • Operations

  • Building

  • Integrations

  • Products

    Powerful modeling and versatile presentation of your entire catalog.

  • Subscriptions

    Sell recurring physical and virtual products alongside one-time offerings.

  • Discounts

    Get the sale with coupons, BXGY promotions, and automatic discounts.

  • Wholesale

    Sell B2B like it's DTC, along with volume pricing, customer groups, and invoicing.

  • Content

    Manage all your products content through the admin dashboard.

  • Users

    Multi-store admin accounts and role-based permission controls.

  • Customers

    Manage customer info, generate reports, and see buyer activity.

  • Orders

    Edit orders anytime and get the right information for smooth fulfillment.

  • Fulfillment

    Ship from multiple locations, track inventory, and split shipments.

  • Reporting

    Monitor your store's performance to ensure you have visibility across the business.

  • Storefronts

    Swell storefronts are fully customizable, allowing you to create just the right experience.

  • Checkouts

    Use our hosted checkout, integrate with a partner, or build a custom flow.

  • Payments

    Connect multiple gateways simultaneously, store cards, and split payments.

  • Internationalization

    Go global with region-specific languages, pricing, and payment methods.

No-code integrations

Connect with 40+ services for marketing, payments, fulfillment, automation, and more.

See all integrations →

Use Cases

  • Direct-to-consumer

    Tell your story and give customers a unique shopping experience

  • Subscriptions

    Sell personalized subscription bundles, memberships, and one-time items together

  • B2B/B2C

    Support retail and wholesale customers from one catalog and dashboard

  • Marketplaces

    Create a B2B or B2C marketplace with multi-vendor carts and split payouts

Customer Stories

  • Spinn Coffee

    A coffee revolution sparked by a connected machine and marketplace

  • Smashing magazine

    Global tax and shipping for complex product bundles

  • Infinitas Learning

    Delievering leading educational experiences in Europe

All customer stories →

Documentation

  • Quickstart

  • Backend API reference

  • Frontend API reference

  • Guides

  • Core concepts

  • Storefronts

  • Headless commerce

Community

  • GitHub

  • Discussion forum

  • Changelog

  • API status

Resources

  • Help Center

    The latest industry news, updates and info.

  • Customer stories

    Learn how our customers are making big changes.

  • Become a partner

    For agencies creating innovative commerce experiences.

Latest blog posts

  • deep-dive-app-functions
    Nov 06, 2025

    Build smarter workflows with App Functions

  • storefronts-v2-and-the-future-of-swell-apps
    Oct 22, 2025

    Storefronts V2 and the future of Swell Apps

  • Changelog

  • API Status

  • Contact us

  • Swell Merch

Blog

AI Agent Orders, Fraud and Chargebacks: A 2026 Merchant Guide

How to spot, approve and defend AI agent orders in 2026: agent signals at payment time, liability by dispute type, rule fixes and the evidence to keep.

Swell Team | November 02, 2026
Hand holding a payment card above a laptop keyboard.

When an AI agent places an order, the card networks still treat it as a card-not-present ecommerce sale, and the merchant still owns the dispute. What changes is the evidence. The signals your fraud stack relies on (a browser session, a human typing, a home IP address) are partly or completely missing, and new signals take their place: a token scoped to an agent, an agent identity flag in the authorization, and a cryptographic signature on the HTTP request.

Most 2026 fraud stacks were tuned to treat fast, scripted, datacenter-hosted checkouts as attacks. That is exactly what a legitimate shopping agent looks like. The result is two failure modes at once: false declines on real agent orders, and weak dispute files when an agent order is later challenged as unauthorized or wrong.

This guide is for whoever owns fraud and disputes. It is organized by where things break in the order lifecycle: identifying the order at payment time, mapping liability per dispute type, fixing rules that misfire, and capturing mandate evidence. For how the agent checkout protocols themselves work, see our Agentic Commerce Protocol explainer.

Key Takeaways

  • Agent orders file under the same reason codes as any other card-not-present order: Visa 10.4, 13.1, 13.2 and 13.3, and Mastercard 4837, 4841, 4853 and 4855. As of October 2026 we found no agent-specific reason code published by either network.
  • Visa's Core Rules (in place since the October 2025 edition and carried into the 18 April 2026 edition) make the cardholder responsible for an agent's actions "as if the Cardholder initiated the Transaction", which turns many "my agent went rogue" claims into evidence fights you can win, not automatic fraud losses.
  • A tokenized agent credential does not shift fraud liability on its own. Under Visa's rules, a 10.4 fraud dispute becomes invalid only when the transaction carries authentication data, such as a 3DS CAVV or a token cryptogram with an approved cardholder verification.
  • Bot challenges, datacenter IP blocks, velocity limits and new-account scoring all misfire on legitimate agents. Allowlist verified signatures, not user-agent strings, and route agent traffic into its own velocity buckets.
  • Per agent order, keep the agent's identity or signature, the token's limits, the confirmation you sent to the human, and timestamps. Store them on the order, not in a log you cannot find six months later.

1. Telling an agent order from a human order and a hostile bot

At payment time you have three places to look. Each reaches your risk stack through a different pipe, and each is at a different stage of maturity. The goal is not to block automation, it is to separate verified, consented automation from everything else.

Signal 1: the credential itself

Agent purchases increasingly arrive on a credential issued for the agent, not the raw card number. With Stripe, the merchant receives a shared payment token that carries `usage_limits` (currency, maximum amount, expiry), plus the card brand and last four digits. Stripe's docs say that once you confirm the payment, refunds and reporting "behave as if you provided the PaymentMethod directly", so downstream it looks like any card payment unless you record where it came from. Stripe also says SPTs can relay Radar risk signals such as the likelihood of a fraudulent dispute, card testing or a stolen card. On the card-network side, agent credentials are network tokens bound to a registered agent. Delegated payment tokens are covered in the ACP explainer linked above, so they are not repeated here.

How it reaches you: through your payment processor's API response and dashboard. Status: live; Stripe's docs list SPTs as available in the US, Canada and select European countries, on a preview API version (checked October 2026).

Signal 2: agent indicators in the authorization data

Visa now defines an "Agentic Transaction" in its Core Rules (18 April 2026 edition): an ecommerce transaction undertaken by a registered Agentic Payment Provider on the cardholder's behalf, based on cardholder-defined instructions, without direct interaction between the cardholder and merchant. The same rules require the agent provider to pass the cardholder's name, billing and shipping address and email to the merchant. That last requirement matters for disputes: you should always have a human email address to send a confirmation to.

Mastercard is adding issuer and merchant-facing signals on top of its agent tokens. On September 30, 2026 it announced trust and intelligence services for Agent Pay: an agent-initiation probability score, now rolling out for testing in the US, plus unusual-activity and review signals, with behavior and credential-risk intelligence planned later. How it reaches you: through your acquirer or processor, if and when they expose these fields. Status: Visa's rule definitions are live; Mastercard's new signals are in testing, so do not build a hard rule on them yet.

Signal 3: signed HTTP requests (Web Bot Auth)

Web Bot Auth lets an agent sign each HTTP request with a private key whose public half is published at a well-known URL, using the `Signature`, `Signature-Input` and `Signature-Agent` headers. The protocol is an active IETF working-group draft dated September 1, 2026, not a finished RFC. Cloudflare already verifies these signatures (its Web Bot Auth docs cover the headers and signing). Cloudflare's Bot Management variables include `cf.bot_management.signed_agent`, which is true when a known agent self-identifies with Web Bot Auth, alongside `cf.bot_management.verified_bot` for allowlisted bots. Since July 1, 2026 Cloudflare classifies signed agents as Verified, split into direct and intermediary agents.

Visa's Trusted Agent Protocol and Mastercard Agent Pay both build on this. Cloudflare's write-up describes separate signature tags for browsing (`agent-browser-auth`) and for paying (`agent-payer-auth`), so a merchant can tell a price-checking visit from a purchase. How it reaches you: at the CDN or WAF layer, before the request touches your checkout. Status: live at Cloudflare's edge; the standard is still a draft.

Putting the three together

A legitimate agent order in late 2026 often shows a verified signature at the edge, a scoped or agent-bound token at payment, and cardholder details passed through by the agent. A hostile bot usually shows none of these, or shows a spoofed user-agent string with no signature behind it. Use that combination to set the order's channel to "agent" at checkout and keep that label on the order record for every later decision.

2. Liability map: who carries the loss for each dispute type

The premise from our ACP explainer holds: the merchant of record owns the dispute. The useful question is which dispute you are facing, because the evidence that wins each one is different.

True third-party fraud

A stolen card or compromised wallet was used through an agent, and the real cardholder never asked for anything. This files as Visa 10.4 (Other Fraud, Card-Absent Environment) or Mastercard 4837 (No Cardholder Authorization), and sometimes 4863 (Cardholder Does Not Recognize, Potential Fraud), per Sift's summary of Mastercard reason codes.

  • Default: merchant loss, as with any card-not-present fraud.
  • Where authentication shifts it: Visa's 10.4 rules list as invalid an ECI 5 transaction authenticated with EMV 3DS and a CAVV, and an ECI 5 transaction on an authenticated payment credential with a token cryptogram (TAVV) and an approved cardholder verification method. If an agent credential arrives with that authentication data, the issuer cannot bring a 10.4.
  • Where it does not: a token alone, or a scoped processor token with no cardholder verification attached, leaves the fraud liability where it was. Ask your processor which of your agent transactions actually carry authentication data; do not assume.

"My agent exceeded my instructions"

The cardholder set up the agent, but says it bought something they did not want or spent more than intended. This is the new category, and issuers will often still code it as fraud (10.4 or 4837) because the cardholder says "I didn't authorize this."

  • Visa's position: rule 4.1.24.10 in Visa's Core Rules (added in the October 2025 edition and carried into the 18 April 2026 edition) says the cardholder "is responsible for any actions taken by an Agentic Payment Provider" as if the cardholder initiated the transaction. The same rules define the agent provider as the cardholder for an Agentic Transaction.
  • What that means for you: you have a strong argument that this is not fraud, but you still need to prove the order came through a registered agent acting on a mandate. Without that proof it looks like any other unauthorized claim.
  • What you cannot do: Visa's rules state a merchant must not require a cardholder to waive the right to dispute a transaction, "including an Agentic Transaction". Your terms can set expectations, not remove dispute rights.
  • Open question: Worldpay's analysis notes that without a signed record of what the shopper authorized, an agent-versus-shopper disagreement "becomes a coin flip." Mastercard's Verifiable Intent framework, open-sourced on March 5, 2026, is meant to create that record; at announcement, integration into Agent Pay was described as coming "in the coming months." Treat it as announced until your processor passes it through.

Wrong item, size or quantity

The agent bought the right kind of thing but the wrong variant or too many units. This files as Visa 13.3 (Not as Described or Defective Merchandise/Services) or Mastercard 4853 (Cardholder Dispute). If the shopper returned it and you did not credit it in time, it can become 13.6 (Credit Not Processed) or 13.7 (Cancelled Merchandise/Services).

  • Default: the merchant wins if the product matched its description and the policy was disclosed; the merchant loses if the listing or variant data was ambiguous.
  • Agent twist: your product data is now the "description" the agent relied on. Clean variant names, units and pack sizes are dispute evidence, not just merchandising. Our guide to preparing a product catalog for AI shopping agents covers the data side.

Item not received

Visa 13.1 (Merchandise/Services Not Received) or Mastercard 4855 (Goods or Services Not Provided). Nothing about this changes for agent orders except one thing: the agent may have entered a shipping address the human did not expect, such as an office or a relative. Delivery confirmation to the address the agent submitted, plus proof that the address came from the agent's request, is your defense.

Repeated agent purchases

Visa now defines an "Agentic Repeated Transaction", where the cardholder consents to the agent placing future orders. If the human forgets the instruction, expect Visa 13.2 (Cancelled Recurring Transaction) or Mastercard 4841 (Canceled Recurring or Digital Goods Transactions). Visa requires the agent provider to get the cardholder's acknowledgement that repeated orders continue until the cardholder changes or cancels the instruction, or it expires.

Is there an agent reason code?

No. The dispute conditions in Visa's 18 April 2026 rules are unchanged in structure; the only agent-specific wording we found in the 10.4 section is that login IDs "for an Agentic Payment Provider" count as matching data elements in the Compelling Evidence 3.0 remedy. We found no Mastercard agent-specific reason code in public material as of October 2026. Separately, Visa's rules expand the Compelling Evidence 3.0 remedy effective 24 October 2026, allowing prior undisputed transactions at one or more merchants to qualify (see our BFCM high-risk guide). For the general Compelling Evidence 3.0 rules, see our high-risk BFCM guide (linked in section 5); the agent change is simply that the agent platform's account ID becomes a usable match field.

3. Why current fraud rules misfire on legitimate agents

HUMAN Security's 2026 State of AI Traffic report found agentic AI traffic grew 7,851% year over year in 2025 (from a low base) and that 2.3% of that traffic reached checkout pages. The same report says only half of one percent separates the rate of benign automation from malicious automation across the interactions it analyzed. In other words, behavior alone no longer tells you intent. Here is where common rules break.

WAF challenges and CAPTCHA

A JavaScript challenge or CAPTCHA at cart or checkout stops a shopping agent cold. An agent that hits one either fails or has to hand control back to the human, who may never come back. The challenge is doing its job against scrapers and failing against customers.

Headless-browser and datacenter-IP blocks

Many agents run in cloud browsers (several of the first signed agents Cloudflare named in August 2025, such as Browserbase and Anchor Browser, are cloud browser services), so they present headless or automation-flavored fingerprints and come from datacenter IP ranges. Rules that block or heavily score "hosting provider ASN" or "automation detected" will catch them.

Velocity limits

An agent comparing options may add and remove items quickly, check several shipping methods, and complete checkout in seconds. Separately, one agent platform's egress IPs carry thousands of unrelated shoppers. Per-IP and per-session velocity rules see one very busy, very fast customer.

Shipping, billing and IP mismatches

The agent's IP reflects the agent's server, not the shopper's location, so IP-to-billing distance is meaningless. Agents are also often used for gifts and errands, so shipping-billing mismatches are more common, not less.

New-account and guest scoring

An agent typically checks out as a guest or creates a fresh account at your store. Rules that penalize account age under one day, first-time devices and no prior orders score every agent order as high risk.

A concrete false-decline scenario

Here is an illustrative case. A shopper in Denver asks her agent to order a $240 pair of trail shoes in size 8 for her sister in Ohio. The agent opens a cloud browser in Virginia, passes your product page, and creates a guest checkout. Your stack sees: a datacenter IP 1,500 miles from the billing address, a shipping address in a third state, a two-second form fill, a brand-new account, and the same card attempted at two other stores in the previous ten minutes (the agent was comparing stock). Each signal adds points; together they cross your decline threshold. The issuer never sees the order, the shopper gets a vague failure message from her agent, and the agent buys from the next store on its list. Nothing in your reports shows a lost customer, only a declined order that looks like a successful block.

Signifyd, citing its own research, says 27% of wrongly declined customers never return. With agents, the lost customer may not even know your store was the one that refused.

4. Checklist: rule changes for agent traffic

These changes are about risk decisions, not integration. Make them in your WAF, your fraud tool and your order workflow. If you have not baselined your current dispute rate, the chargeback figures in our high-risk payment gateway statistics give useful context.

  1. Allowlist verified signatures, not user-agent strings. A user-agent string is free to fake. Create a WAF rule that skips bot challenges on cart and checkout paths only when the request carries a verified signature (on Cloudflare, the `signed_agent` or `verified_bot` fields). Keep challenges in place for everything unsigned.
  2. Separate "browse" trust from "pay" trust. Where the signature tag distinguishes browsing from paying, allow verified browsing broadly, but require the payment tag plus an agent-bound or scoped credential before you treat the checkout as an agent order.
  3. Give agents their own velocity buckets. Key velocity on the verified agent identity plus the shopper's email or token, not the IP. Set limits per shopper within an agent, so one compromised shopper cannot ride the agent platform's reputation.
  4. Drop or reweight IP-geography features for verified agent orders. IP-to-billing distance is noise when the IP belongs to the agent. Keep address verification and email-age checks, which still describe the human.
  5. Step up instead of declining. For agent orders that score borderline, hold the order and send the human a confirmation link rather than declining. A hold costs you hours; a decline loses the sale and teaches the agent to skip you.
  6. Use the token's own limits as a rule input. If the order total is close to the token's maximum amount, or the token expires within minutes, that is normal for agents. A total above the stated maximum should never authorize; if it does, investigate the integration, not the shopper.
  7. Cap high-value and resellable SKUs. Electronics, gift cards, limited sneakers and similar items are where stolen-credential agents will go. Set per-order quantity limits and value thresholds that route agent orders to review for those SKUs specifically, instead of tightening the whole catalog.
  8. Write Radar-style rules on agent metadata. If you use Stripe and your Radar plan supports custom rules, Radar rules can evaluate metadata you attach to the payment (for example `::channel:: = 'agent'`). Tag agent orders at checkout so you can review, allow or block that segment separately and report its dispute rate on its own.
  9. Measure false declines on the agent segment. Track approval rate, manual-review rate and dispute rate for orders tagged as agent orders versus human orders. Without the tag you cannot tell whether a rule change helped.

5. Mandate and consent evidence for every agent order

Your generic evidence (delivery proof, AVS and CVV results, policy acceptance) still applies; our BFCM high-risk guide covers it. This list is only what is new for agent orders. Capture it at checkout and store it on the order record, because pulling it from CDN logs months later rarely works.

What to store per agent order

  • Agent identity: the `Signature-Agent` URL or key ID from the verified request, and the agent platform's name. This is what proves a registered agent, not an unknown script, placed the order.
  • Agent platform account or login ID: Visa's 10.4 remedy now accepts login IDs for an Agentic Payment Provider as a matching element across prior undisputed transactions.
  • Credential scope: the token ID, its maximum amount, currency and expiry, and any merchant restriction it carried. An order inside those limits is evidence the human set a budget that covered it.
  • Mandate reference: any intent or mandate ID the agent or processor passes, such as a Verifiable Intent record once your processor supports it.
  • Policy disclosure: the version of your refund and cancellation policy that was available to the agent at checkout. Visa's rules let the agent provider accept a limited cancellation, exchange or refund policy on the cardholder's behalf, either at final checkout or when the cardholder set the instruction, so your policy must be retrievable by the agent, not only shown in a modal.
  • Human confirmation: the confirmation email you sent to the cardholder's email address, with its send time and delivery status.
  • Timestamps: signature creation time, order creation time, authorization time and confirmation send time, in one time zone.

Order confirmation copy for agent-placed orders

Visa already requires the agent provider to make an order confirmation available to the cardholder for at least 120 days, including the merchant's cancellation and refund policies. Send your own as well. The goal is that a surprised human finds a cancel button before they find their bank's dispute form.

Example subject line: "Your order #10482 was placed by your shopping assistant". Example body copy: "Your AI assistant placed this order on your behalf at 2:14 pm MT using your card ending 4417. If you did not intend this purchase, cancel it free of charge before it ships: [Cancel order]. After shipping, you can return it within 30 days."

  • Name the agent or platform if you know it. "Placed by your assistant" is clearer to a human than a bare order number.
  • Put the cancel link above the fold, and honor it automatically until fulfillment starts.
  • Repeat the size, color and quantity in plain text. Wrong-variant disputes start when the human first sees what was bought.

Refund-policy language

Add a short clause covering agent orders, for example: "Orders placed by an AI assistant or other software acting on your behalf are treated as orders placed by you. You can cancel any such order free of charge before it ships using the link in your confirmation email, and standard returns apply after delivery." Keep it consistent with Visa's position that the cardholder is responsible for the agent's actions, and do not phrase it as a waiver of dispute rights, which the rules prohibit. Have counsel review the final wording.

Detailed return and RMA workflows are a separate topic; the point here is narrower: a visible, easy cancel window is the cheapest dispute prevention you have for agent orders.

Reality check: payments, compliance and costs

  • Liability rules are still moving. Visa has written agent responsibilities into its rules; Mastercard's mandate record (Verifiable Intent) and new trust signals were announced or in testing as of October 2026. Expect issuers to apply judgment case by case until dispute outcomes accumulate.
  • Your browser-side tools see less. The ACP explainer notes that fingerprinting and 3DS often do not run in delegated checkouts. The deeper consequence is that your dispute file must lean on server-side records (signature, token scope, confirmation) rather than device data, which is why the evidence list above lives on the order.
  • Agent disputes count like any other. They add to your dispute ratio under network monitoring programs. The thresholds are covered in our BFCM high-risk guide; the practical point is that a spike in agent-order disputes can put the whole merchant account at risk, not just the agent channel.
  • Pilots are pilots. Visa's Intelligent Commerce Connect, a single integration for agent payments across several protocols, was in pilot with seven partners as of April 9, 2026. Do not design rules that depend on fields your acquirer does not yet send.
  • Costs: most of these changes are configuration in tools you already pay for (WAF, fraud screening, email). The real cost is analyst time to review a new segment for the first few months.

How to choose which changes to make first

Pick based on what you see today, not on what the networks have announced.

  • You see agent traffic but few agent orders: start with section 3. Your WAF or fraud tool is probably blocking them. Allowlist verified signatures on checkout paths and watch approval rate.
  • You see agent orders and rising "unauthorized" disputes: start with section 5. Tag orders, send the confirmation email with a cancel window, and store the agent evidence fields.
  • You sell resellable or high-ticket goods: do both, and add SKU-level caps for agent orders before loosening anything else.
  • You do not see agent traffic yet: add the order tag and evidence fields now. They cost little, and you will want the history when volume arrives.

Whatever you choose, the backend has to let you attach this data to the order and act on it. On an API-first platform like Swell, orders include a `metadata` field for custom checkout values, and you can add custom fields to built-in models such as orders, so agent identity, token limits and confirmation timestamps live on the order itself. Webhooks on events such as `order.created` can send each agent order to your fraud tool, and the order's `hold` status gives you a place to park borderline orders for review instead of declining them. Rules can then be applied to the agent segment on its own, since the segment is a field you control. Swell's published plans include annual revenue up to a set ceiling ($50K on Starter at $29 per month billed yearly, up to $5M on Unlimited at $2,250), with an overage rate above the ceiling that falls as you upgrade, from 2% to 0.4%. If you are evaluating where agent orders will come from, our overview of agentic commerce platforms and custom checkout statistics are useful next reads.

Frequently Asked Questions

Who pays when an AI agent order turns out to be fraud?

Usually the merchant, exactly as with any card-not-present fraud. The exception is when the transaction carries authentication data, such as a 3DS CAVV or a token cryptogram with an approved cardholder verification; Visa's rules then make a 10.4 fraud dispute invalid. Ask your processor which agent transactions include that data.

Are there new chargeback reason codes for AI agent purchases?

Not that we could find as of October 2026. Agent disputes file under existing codes, most often Visa 10.4, 13.1, 13.2 and 13.3, and Mastercard 4837, 4841, 4853 and 4855. Visa's April 2026 rules do add agent wording to the Compelling Evidence 3.0 remedy, allowing an agent platform's login ID as a matching element.

What if the customer says their agent bought something they didn't want?

Under Visa's current Core Rules (18 April 2026 edition), the cardholder is responsible for an agent provider's actions as if they had initiated the transaction. That gives you grounds to contest, but you need evidence: the agent's verified identity, the token's limits, and a confirmation you sent to the human with a cancel option they did not use.

Should I block AI agents at checkout?

Block unverified automation, not agents as a category. Allow requests that carry a verified Web Bot Auth signature through your bot challenges on checkout paths, and keep challenges for everything else. Blocking all agents means losing orders you cannot see in your reports.

Does Web Bot Auth prove the shopper approved the purchase?

No. It proves which agent sent the request. Consent comes from the payment side: the scoped or agent-bound token, any mandate record the agent passes, and your own confirmation to the human. You need both for a strong dispute file.

What should an order confirmation for an agent order include?

That the order was placed by an assistant on the customer's behalf, the item, variant and quantity in plain words, the card's last four digits, the total, and a one-click cancel link that works until the order ships. Send it to the email address the agent passed at checkout.

Do I need new fraud software for agent orders?

Not necessarily. Most of the changes in this guide are configuration: WAF rules keyed on verified signatures, velocity rules keyed on agent and shopper identity, a hold-for-review path, and order metadata your fraud tool can read. Add tooling only if your current stack cannot read those fields.

Build your store on Swell

Headless commerce with a full API on every plan, native subscriptions and custom data models. Start a store and see how it fits your build.

Try Swell for freeTalk to us

Next-level commerce for everyone.

X.comGitHubLinkedIn

Subscribe to our newsletter for product updates and stories

Resources

Help CenterDeveloper CenterCommunityChangelogLearnSwell Merch

Use cases

SubscriptionsB2B WholesaleMarketplaceOmnichannelDirect-to-consumerEnterprise

Explore

FeaturesPricingIntegrationsCustomer stories

Developers

OverviewDocumentationGuidesStorefrontsHeadlessSwell Apps

Company

BlogAbout usPartnersContact us

© 2026 Swell. All rights reserved.

Privacy PolicyTerms of Service