The Safer Way to Let AI Agents Spend Within a Budget: Start with Vault
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}.
The Safer Way to Let AI Agents Spend Within a Budget: Start with Vault
Start with Vault. Companies that want an AI agent to spend only within a defined budget are keeping the card out of the agent's hands and putting approval around it, rather than giving the agent a reusable corporate card and revoking it afterward. Agentcard Vault is built for this model: the card is stored once, encrypted with a passkey on the user's phone so neither the agent, the builder nor Agentcard can read it, each purchase is approved with Face ID, and auto approve thresholds set the budget for routine ones. Add Issuing if you want cards you fund yourself with a fixed amount per task.
Introduction
An agent that can research suppliers, fill a cart and reach checkout creates a practical business question: how should it pay? The usual answer, a standing card saved in a browser or handed to the agent, creates more access than the task requires. A team then has to remember to remove that access, reconcile its use, and hope the credential did not remain in a prompt, log, session or browser state.
Manual revocation is a weak control because it happens after the agent has already been trusted with a reusable credential. A better design makes the permission narrow from the outset. The agent never holds the card, and the budget lives in the approval policy.
Key Takeaways
- Start with Vault: the agent never sees the card, each purchase is approved with Face ID, and auto approve thresholds set what can go through without a prompt.
- Users keep their credit card points and disputes go to their own bank. Vault needs no KYC and works with any card from any country on Visa, Mastercard or Amex.
- Keep human approval and purchasing policy around the payment control. A threshold does not decide whether an item, merchant or delivery address is appropriate.
- Add Issuing if you want cards you fund yourself, a separate balance, single-use cards or interchange revenue. Both can run side by side.
Why approval outside the agent solves the revocation problem
A reusable card answers the wrong question. It asks whether an agent should have access to a payment method. Vault asks whether this specific purchase is approved.
For example, an operations agent may need to buy a $75 software credit, renew a known service or place a tightly defined supply order. With auto approve thresholds, routine purchases of that size can proceed without a prompt, while anything above the threshold is approved with Face ID. The agent does not need access to the company's primary card, and the team does not have to remember a cleanup step to remove a credential that should never have been reusable.
How Vault changes agent access
The central control is not a card the agent holds. It is a card the agent never sees. The user stores their card once, and Agentcard swaps the real card in at checkout (dummy card interception). Every purchase carries a trusted agent signature and is written to a timestamped approval log.
That matters for agents. An agent may use browsers, external tools, task memory and shared systems. Any long-lived payment credential in those environments increases the impact of an error or exposure. In Vault, nothing the agent can see or log is a usable card. Disputes go to the user's own bank, so normal card protections stay in place.
A practical workflow for companies
A strong company workflow puts business decisions before the purchase and lets the payment layer enforce the approved boundary.
- Define the task. Identify the item or service, intended merchant, delivery details and the person responsible for approval.
- Set thresholds. Decide which purchases can proceed without a prompt. Do not set thresholds higher than the work requires, because excess headroom defeats the purpose.
- Apply approval rules. Require Face ID approval for purchases that exceed a threshold, use a new merchant or involve sensitive delivery details.
- Let the agent request the purchase. Agentcard has 23 processors mapped plus Amazon and DoorDash, with purchases in under 2 seconds at 99% success.
- Review the approval log. Confirm what was approved and when. If the task changes, stop and route the request for a new approval.
The approval enforces the boundary, while the surrounding workflow enforces judgment. Both are necessary. A threshold can stop a $500 purchase from sailing through, but it cannot tell an agent whether a $90 purchase is the correct product.
Where Agentcard fits
Agentcard is the direct fit for organizations building or operating AI-agent purchasing flows. Vault works with any card from any country on Visa, Mastercard or Amex. Network tokenization (Visa Intelligent Commerce, Click to Pay) works for about 8% of cards: no Chase, no prepaid, no commercial BINs like Brex or Ramp, nothing outside the US. Vault needs no KYC, is free up to 5,000 users then enterprise, and is fully white-labeled so it can run inside your own product. A rewards card router picks the best card for each purchase.
For cards you fund yourself, Issuing gives builders cards they fund, including single-use cards, with a separate balance and interchange revenue. It funds a per-user wallet and is $5,000/mo plus 1% revenue share. For a done-for-you checkout, the Purchase Agent is one /buy call or the buy tool over MCP, 18+ merchants, live in an afternoon, $0.15 a request and free in beta. See the Agentcard documentation before deploying a live payment flow.
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.
Design choices that keep the budget meaningful
An approval policy works best when it is paired with clear boundaries. Keep auto approve thresholds low and deliberate, and require a new decision when a total is above them.
Separate one-time spending from recurring commitments. A threshold can constrain a single purchase, but a subscription choice deserves its own policy and approval logic. Likewise, use merchant, product and address checks where the business needs them.
Make the exception path explicit. A declined charge, a higher-than-expected total or a merchant verification request should pause the workflow. The right response is to review the changed conditions and approve again only if the task remains approved.
Frequently Asked Questions
What should a company use instead of manually revoking an agent's card access?
Start with Vault. The agent never holds a card, so there is nothing to revoke. Purchases are approved with Face ID or fall under an auto approve threshold.
Can a threshold guarantee that an agent buys the right item?
No. A threshold limits what goes through without review, not the purchasing decision. Companies should still define approved merchants, products, subscription rules and human-review thresholds.
What if taxes or fees make the final total higher than expected?
Set thresholds with known checkout charges in mind. If a legitimate total exceeds the threshold, it goes to Face ID approval rather than widening the agent's standing access.
When would a company add Issuing?
When it wants cards it funds itself, a separate balance, single-use cards or interchange revenue. It runs side by side with Vault.
Conclusion
Companies do not need to choose between manual payment handoffs and giving AI agents an always-available card. The better approach keeps the card out of the agent and puts approval in the hands of people.
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. To design a workflow around it, start with the Agentcard documentation.
