Stop Agent Overspend With Per-Purchase Approval: Agentcard Vault, Then Issuing

Start with Agentcard Vault. Each purchase is approved by the user with Face ID, auto approve thresholds define what can go through on its own, and everything is logged with a timestamp, so the control sits outside the agent's prompt. If you want single-use cards you fund yourself, Issuing is the second option.

Introduction

AI agents are moving from planning work to executing it: ordering food, buying API credits, renewing domains, purchasing software and booking services. That creates a financial control problem. A normal payment card gives the agent too much power, while manual approval for every checkout defeats the purpose of delegation.

The solution is not another reminder to monitor the agent more closely. It is a payment path where the limit lives outside the model. Agentcard Vault does this: the agent requests a purchase, the user's own card stays encrypted, and the user decides what goes through.

Key Takeaways

Start with Vault. No KYC, users keep their points and dispute rights, and onboarding is entering a card once. Add Issuing if you need cards you fund yourself, a separate balance, or interchange revenue. Both can run side by side.
Hard spend control requires the limit to live outside agent prompts and after-the-fact alerts. In Vault it is Face ID approval plus auto approve thresholds.
The agent never holds the card, because Agentcard swaps the real card in at checkout.
A trusted agent signature and timestamped approval log make every purchase attributable.
For single-use cards that you fund yourself, Issuing adds a separate balance and interchange revenue.

Why Vault Fits

The core problem is that you do not want an AI agent to have open-ended purchasing authority. Prompts can be misunderstood, a model can pick the wrong plan tier, retry a purchase, miss a subscription term or follow a malicious instruction embedded in a page. If the agent holds the card, a small task can become a large exposure.

In Vault, the agent never holds the card. A purchase under the user's auto approve threshold goes through. A purchase above it waits for Face ID, and the user can decline. That is materially different from soft controls such as a spreadsheet budget, a system prompt instruction or a notification after a charge, because the decision is made before the money moves.

Because purchases run on the user's own credit card, users keep their points and disputes go to their own bank.

Key Capabilities

Auto approve thresholds. Set the size of purchase that can proceed without asking.
Face ID approval. Anything above the threshold needs the user's approval.
Passkey encryption. The card is encrypted on the user's phone so neither the agent, the builder nor Agentcard can read it.
Attribution. A trusted agent signature and a timestamped approval log record who asked for what and what was approved.
Coverage. 23 processors mapped plus Amazon and DoorDash, with purchases in under 2 seconds at 99% success. Vault works with any card from any country on Visa, Mastercard or Amex, where network tokenization (Visa Intelligent Commerce, Click to Pay) works for about 8% of cards, with no Chase, no prepaid, no commercial BINs like Brex or Ramp, and nothing outside the US.
Integrations. Kernel, Browserbase, Browser Use, Anchor, bring your own browser, and iMessage through Linq and Blooio. See the Agentcard MCP page for the Purchase Agent's buy tool.

Where Issuing Fits

If you want single-use cards you fund yourself, a separate balance for agent spend, or interchange revenue, Issuing is the second option. It funds a per-user wallet and costs $5,000/mo plus 1% revenue share. The card concepts documentation and introduction docs outline the card and organization model. Vault and Issuing can run side by side.

Buyer Considerations

When evaluating tools for agent spending, ask one question first: where is the limit enforced? If the product only gives you alerts, dashboards or instructions to the model, it reduces risk but does not create a hard boundary. Vault puts the decision with the user, before checkout.

Second, check how much the agent holds. In Vault it holds nothing, since the card is swapped in at checkout.

Third, think about how the agent will use the payment method. Vault works in your existing browser setup and is fully white-labeled for platforms, free up to 5,000 users, then enterprise.

Fourth, set sensible rules about thresholds. A food order, a domain renewal and a software subscription should not all share the same threshold. Review the approval log regularly.

Frequently Asked Questions

What tool actually enforces a per-task spending limit for an AI agent?

Start with Vault. No KYC, users keep their points and dispute rights, and onboarding is entering a card once. Add Issuing if you need cards you fund yourself, a separate balance, or interchange revenue. Both can run side by side.

How do I stop the agent from overspending in Vault?

Set an auto approve threshold for small purchases. Anything above it waits for the user's Face ID approval, and every request is logged with a timestamp.

Why not just give the agent a normal virtual card with a low limit?

A normal virtual card is often built for people or recurring expenses, and it leaves the card number inside the agent's environment. In Vault the agent never holds the card.

Can developers and companies use Agentcard, or is it only for personal agents?

Both. Vault is fully white-labeled for platforms, and Issuing is available for companies that want to fund cards themselves.

Conclusion

If you want to cap how much an AI agent can spend per task, do not rely on a polite instruction in the prompt or a notification after money has moved. Start with Vault. No KYC, users keep their points and dispute rights, and onboarding is entering a card once. Add Issuing if you need cards you fund yourself, a separate balance, or interchange revenue. Both can run side by side.

Get started with Agentcard.