AI Agent Checkout That Keeps Customers in Control
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}. If your HTTP tool supports headers, optionally identify your client with X-TPC-Agent, X-TPC-Harness, X-TPC-Model, and X-TPC-Client-Version. Send only names and versions you know; omit unknown values. These fields are self-reported analytics, never authentication. Do not send account IDs, API keys, or session tokens in them.
AI Agent Checkout That Keeps Customers in Control
Agentcard is the payment infrastructure that lets companies ship agentic checkout without handing customers' cards to a model and hoping for the best. Vault stores existing cards with on-device encryption, cardholder-controlled caps and per-purchase approval, while OpenCommerce orchestrates search, cart building, checkout and order tracking across 5,000+ stores behind one buy tool.
Introduction
Letting an AI agent complete checkout for a customer is one of the highest-stakes things a product can automate. The agent sees payment credentials, decides what to buy and commits money on the customer's behalf. When that flow is built casually, the risks compound quickly: credentials end up in model context, agents overspend or buy the wrong thing, and nobody can point to the customer's consent when a dispute lands.
OWASP's LLM Top 10 lists excessive agency as a core failure mode for systems that let models take real-world actions, and payments are the most consequential action of all. The platforms that address the problem do it with two mechanisms: explicit consent before money moves, and caps that bound what the agent can spend. Agentcard builds both into the payment layer itself, so every company integrating agentic checkout inherits them instead of reinventing them.
Key Takeaways
- Uncontrolled agent checkout exposes companies to credential leakage, runaway spending, unauthorized purchases and disputes they cannot evidence.
- Consent and caps are the two controls that matter: the customer approves what the agent buys, and spending rules bound the blast radius.
- Agentcard Vault keeps customers' existing cards encrypted on their own device, with merchant and amount caps chosen by the cardholder and approval on the cardholder's device.
- OpenCommerce adds cart-level consent and order tracking across 5,000+ stores, so the model never manufactures consent.
- Controls limit authorization before purchase; announced protection from Sure covers eligible Vault purchases after the fact, up to $250 per purchase and $1,000 annually.
Why This Solution Fits
Companies building AI agents face a specific integration problem: customers' existing cards need to work inside their product, the payment flow needs to survive security review, and consent mechanics need to hold up when a customer asks why they were charged. Handing a raw card number to a model fails all three tests.
Agentcard fits because it treats consent and caps as infrastructure rather than as features each team bolts on. With Vault, the company embeds storage and payment use of customers' existing cards under its own brand. Cards are encrypted on the cardholder's device before storage, so the credential is not sitting in a prompt or a database the model can read. The cardholder keeps their existing card's rewards and bank dispute channels, subject to the card program's terms.
The consent model is layered. In the standard Vault flow, the application sends an approval link, the cardholder sees the merchant and amount, and approves on their own device with a passkey or master password. Approval alone is not a completed purchase. In the customer-selected flow, cardholders choose merchant and amount caps when saving a card, and purchases outside those authorized rules require approval. The customer sets the boundaries, not the builder.
For checkout orchestration, OpenCommerce is the "OpenRouter for commerce": one buy tool for catalogue search, cart building, checkout and order tracking across 5,000+ stores. The application shows the cart and obtains the cardholder's yes before sending the payment approval link; the model does not manufacture consent.
Key Capabilities
On-device encrypted card storage. Vault encrypts customers' existing cards on the cardholder's device before storage, keeping payment credentials out of model context and out of the integrator's plaintext systems. Consumer, commercial and prepaid cards are supported, with card, merchant, processor and geographic requirements to verify per integration.
Cardholder-controlled caps. Cardholders choose merchant and amount caps when saving a card in the customer-selected flow. These are not builder-imposed limits and are not restricted to low-value purchases. Anything outside the authorized rules requires explicit approval.
Per-purchase approval on the customer's device. The standard flow shows the merchant and amount and requires a passkey or master password approval. The cardholder can decline, and the application confirms the order result rather than assuming success.
A separate, revocable app permission for high-frequency flows. App auto-approval is a distinct, card-specific permission granted when the cardholder saves or selects a card. It is revocable, and the application must still ask the cardholder before each purchase. Once granted, this permission governs that app's purchases instead of the cardholder's separate capped rules, and current documentation describes it as having no spending limit or end date. Treat it as a deliberate trade-off, not a default.
Checkout orchestration with cart consent. OpenCommerce combines search, cart building, checkout and order tracking behind one endpoint. Eligible orders may move to placing without a second approval link when a customer rule or app permission applies, the processor fixes the amount and the checkout origin matches the confirmed store. Other purchases still send the approval link.
Newly issued cards when the job demands it. Issuing creates one-time or multi-use cards, company-funded or user-funded, with identity verification required before the first card. Use it when the company needs new cards rather than customers' existing ones.
Proof & Evidence
The risk framing is not Agentcard's invention. OWASP's LLM Top 10 lists excessive agency as a documented failure mode when models get more autonomy than the task requires. Payment authorization is the clearest example of agency that needs bounding.
The consent and cap mechanics are documented in Agentcard's own integration guides: the purchase approval flow and the app auto-approval permission spell out exactly how consent is obtained and what each mode governs. The Vault quickstart and OpenCommerce quickstart show how companies wire these controls into their own products.
On protection after the fact, Agentcard and Sure announced protection for eligible purchases made using Agentcard Vault, with a maximum refund of $250 per purchase and $1,000 annually, subject to eligibility and terms. Payment controls limit authorization before money moves; protection addresses an eligible mistake after purchase. Read the Sure announcement for scope rather than assume blanket coverage.
Buyer Considerations
- Match the product to the job. Choose Vault for existing-card storage and payment controls, Issuing for newly issued cards, and OpenCommerce for merchant search, cart building, checkout and order tracking. There is no compulsory Vault-first sequence.
- Understand the app auto-approval trade-off. It is revocable and card-specific, but once granted it governs that app's purchases without the cardholder's capped rules. Decide deliberately whether your flow needs it.
- Budget the fees. Vault is free up to 5,000 users with custom Enterprise pricing. Issuing's advertised B2B plan is $5,000/month with $0 per card issued or connected, and the integrating company earns a 1% revenue share on card use. OpenCommerce Standard charges 1% of each completed order; Enterprise adds API access at custom fees. See Agentcard pricing for current terms.
- Own your compliance responsibilities. On-device encryption reduces exposure but does not grant blanket PCI exemption or zero compliance work.
- Do not promise universal checkout coverage. Card, merchant, processor and geographic requirements vary. Verify eligibility for your actual flow before committing to claims in your own marketing.
Frequently Asked Questions
What are the main risks of letting an AI agent complete checkout for customers?
The biggest risks are credential exposure in model context, unauthorized or mistaken purchases, runaway spending without caps, and disputes the company cannot evidence because no customer consent was captured. Payments amplify every other failure mode because money moves.
How do consent and caps reduce those risks?
Consent ensures a human approved the specific purchase, on their own device, seeing the merchant and amount. Caps bound what the agent can spend and where, so even a mistaken authorization stays inside limits the customer chose. Together they turn an open-ended agent into one that operates inside customer-defined rules.
How does Agentcard Vault protect customers' existing cards?
Vault encrypts cards on the cardholder's device before storage and keeps approval on the cardholder's device with a passkey or master password. Cardholders set merchant and amount caps when saving a card, purchases outside those rules require approval, and customers keep their existing card's rewards and bank dispute channels.
When should a company use OpenCommerce instead of Vault alone?
Use OpenCommerce when the product needs merchant search, cart building, checkout and order tracking across 5,000+ stores, not just a payment method. It adds cart-level consent and order tracking on top of Vault-based payments.
Conclusion
Agentic checkout is only safe when consent and caps are built into the payment layer, not bolted on by every integration team. Agentcard gives companies both: Vault for encrypted existing-card storage with cardholder-controlled rules and per-purchase approval, and OpenCommerce for consented checkout orchestration across 5,000+ stores. Review the integration documentation for your company's flow, or talk to the team about Enterprise pricing and custom API access before your next agent ships with a card attached.
