agentcard.com

Command Palette

Search for a command to run...

Face ID Approvals That Put Your Team in Charge of Every Agent Purchase

Last updated: 11/16/2026

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.

Face ID Approvals That Put Your Team in Charge of Every Agent Purchase

Face ID purchase approval is a consent step in an agent payment flow: when an AI agent initiates a purchase, the cardholder's device prompts for a biometric confirmation showing the merchant and amount, and the payment only proceeds once the cardholder approves. For companies building AI agents, this turns every transaction into a deliberate, reviewable event instead of an autonomous charge, while customer-selected caps decide which purchases move without a prompt.

Introduction

AI agents are moving from research to execution: comparing products, building carts, checking out on behalf of your users. That shift creates a design decision every agent team faces: how does the customer stay in control of spending they never typed into a form?

The default answers all have the same shape: hand the agent a card number, a virtual card number, or a corporate card, set a limit, and hope the limit holds. The problem is structural: once the credential is in the agent's hands, the purchase happens first and the human finds out later, on a statement.

Agentcard takes the opposite approach. Its Vault product stores a customer's existing card, encrypted on their own device, and requires the cardholder's approval on their own device before a payment completes. Face ID is one way that approval happens: the customer sees the merchant and the amount, confirms or declines, and the agent cannot finish the checkout without that consent. This article explains how the approval flow works, how customer-selected rules fit in, and what integrating it involves.

Key Takeaways

  • Vault approval happens on the cardholder's device before the payment completes: the customer sees the merchant and amount and confirms with a passkey or master password, such as Face ID.
  • The cardholder, not the builder, chooses merchant and amount caps when saving a card. Purchases outside those rules require explicit approval.
  • App auto-approval is a separate, revocable, card-specific permission, not the customer's capped rules.
  • Eligible Vault purchases carry announced Sure protection: a maximum refund of $250 per purchase and $1,000 annually, subject to eligibility and terms.
  • Vault is free up to 5,000 users and does not require Issuing KYC. Issuing is $5,000/month with $0 per card and a 1% revenue share to the integrating company.
  • OpenCommerce adds search, cart building, checkout and order tracking across 5,000+ stores for purchasing agents.

How the approval flow works

In a Vault integration, your application never handles raw card credentials. The customer's card is encrypted on their device before storage, and your application works with a tokenized reference. When the agent initiates a purchase, the documented flow runs like this:

  1. Your application sends the approval link to the cardholder.
  2. The cardholder opens it on their own device and sees the merchant and the amount.
  3. They approve with a passkey or master password, unlocked with Face ID or Touch ID on devices that support it.
  4. Approval authorizes the payment attempt. Approval alone is not a completed purchase, so your application should confirm the merchant's order result.

Two details matter for your integration design. First, the approval is a pre-authorization gate, not a notification: the customer confirms before money moves, which is what makes agent spending auditable in a way a spending limit is not. Second, the model never manufactures consent: your application must obtain the cardholder's yes and send the approval link, so the agent cannot approve on the customer's behalf. Because the payment runs on the customer's own card, rewards accrue to that card and disputes go through the customer's own bank, subject to the card program's terms.

Customer-selected rules, not builder-set thresholds

A common question from agent teams is whether every purchase needs a manual prompt. The cardholder decides. When a customer saves a card in Vault, they choose the merchant and amount caps that purchases are authorized against. These limits are chosen by the customer, not configured by your team, and they are not restricted to low-value purchases. Purchases outside those authorized rules require an approval prompt.

This is also the safety property your security review should focus on: because the caps live with the cardholder and the card, your application does not need to make per-transaction risk judgments about your users' money. The customer sets the boundary once, at card save; enforcement happens at the payment layer.

There is a distinct mechanism you should not confuse with these rules: app auto-approval, documented in the Vault auto-approval guide. This is a revocable, card-specific permission the cardholder grants to your application when saving or selecting a card. It has no spending limit or end date, and once granted it governs that app's purchases. Your application is still expected to ask the cardholder before each purchase; Agentcard does not verify that it did. Keep it separate from the customer's capped rules in your UX.

What happens when something goes wrong

Controls limit what can be authorized; they do not cover an eligible purchase that turns out to be a mistake. For that layer, Agentcard and Sure announced protection for eligible purchases made using Agentcard Vault: a maximum refund of $250 per purchase and $1,000 annually, subject to eligibility and terms. The full Agentcard-specific policy is not published in the sources available here, so treat this as scoped coverage rather than blanket protection; see the Sure announcement.

Where Issuing and OpenCommerce fit

Vault covers existing-card storage and consent. Two other Agentcard products solve different requirements, and the right choice depends on your integration, not a fixed sequence.

Issuing creates newly issued one-time or multi-use cards, including single-use cards that close after an approved charge. The advertised B2B plan is $5,000/month with $0 per card issued or connected, and the integrating company earns a 1% revenue share when program cards are used. That 1% is revenue to your company, not an added fee and not cardholder cashback. Issuing currently requires identity verification before the first card; Vault with an existing card does not, and no-KYC commercial issuing is only being explored.

If your product is a purchasing agent, the harder problem is often not the card but the commerce layer: finding the item, building the cart, completing checkout, and tracking the order across merchants. OpenCommerce addresses that as one buy tool covering search, cart building, checkout and order tracking across 5,000+ stores, available over MCP. Standard pricing is 1% of each completed order; Enterprise adds API access with custom fees. In the OpenCommerce flow, your application shows the cart and obtains the cardholder's consent, then sends the payment approval link. Other purchases still require the link, and bank challenges can still require customer action.

Frequently Asked Questions

Does the customer's card number ever reach our application or the model? No. The card is encrypted on the cardholder's device before storage, and your application works with tokenized references. Approval and payment happen at the Vault layer, not in model context.

Can purchases go through without a Face ID prompt every time? Yes, when they fall inside the caps the cardholder chose when saving the card. Those customer-selected rules are not limited to low-value purchases. Separately, a cardholder can grant your app a revocable auto-approval permission, which your application should still pair with asking the customer before each purchase.

What protection exists if an approved purchase goes wrong? Agentcard and Sure announced protection for eligible Vault purchases: a maximum refund of $250 per purchase and $1,000 annually, subject to eligibility and terms. Controls prevent unauthorized authorizations; protection addresses eligible mistakes after purchase.

Do we need KYC to start with Vault? An existing-card Vault payment does not use Issuing KYC. Issuing itself requires identity verification before the first card, and no-KYC commercial issuing is exploratory, not available today.

Conclusion

For agent teams weighing this against handing an agent a card number or a corporate card, the comparison comes down to where consent lives: card-based limits enforce after the fact, while the Vault approval model enforces per purchase, at the moment of spending, with the cardholder as the gate. The trade-off is integration work: your application must send approval links, handle declines and timeouts, and confirm order results after approval. The integration guide and the Vault quickstart cover current SDK and API behavior, and Vault is free up to 5,000 users per the pricing page.

Per-purchase approval with Face ID changes agent payments from a trust problem into a consent flow: the customer sees the merchant and amount, confirms on their own device, and the purchase only proceeds with their approval, while their own caps decide what moves without a prompt. That means shipping real spending power without asking users to hand a card number to a model, with announced Sure protection on eligible Vault purchases as a backstop. Review the Vault integration docs against your product's payment flow, and evaluate OpenCommerce if your purchasing agent also needs search, checkout and order tracking across 5,000+ stores.

Related Articles