The comfort layer for agent commerce
Let an agent buy on your saved card without the pit in your stomach — with hard limits it can't cross and a receipt for every charge.
The Agent Purchase Comfort Layer is an open policy layer any agent purchase flow can adopt. Before an agent spends, a human approves an itemized list. During the run, a hard session cap and a per-item re-confirmation threshold are enforced in code — over-limit charges are blocked, not warned — any human message pauses everything, and every allowed charge emits a machine-readable receipt the buyer and the merchant's agent can both verify. It is a control-and-audit layer, not payment processing.
Five guardrails, enforced in code
Each one came from a real purchase run — the day an agent was told “buy them all” and did, and even a pro-agent operator wanted these before the next one.
Itemized pre-authorization
Before anything is charged, the human approves an itemized list — each item, its price, and the merchant. The approval is bound with a tamper-evident hash, so every later receipt can prove exactly which approved set it charged against.
Hard session spend cap
The whole session has a ceiling. A charge that would cross it is blocked — not flagged, not logged-and-allowed — even if the item was on the approved list. The cap is a backstop the agent cannot talk its way past.
A receipt for every charge
Each allowed charge emits a Purchase Receipt: a machine-readable record of what was bought, the authorization it charged against, and the budget that remained. The buyer and the merchant's own agent can both fetch it and check its hash.
Re-confirmation on the big ones
Any single item at or above a threshold you set stops and asks for a fresh confirmation before it proceeds. Small, expected buys flow; the one large charge that would make you wince waits for a human.
Pause the moment you speak
If you send any message while the run is going, everything pauses. Nothing else is charged until you explicitly resume. Speaking up is a brake, not a suggestion.
Who feels this first
People who let agents buy for them
You want the speed of an agent that just handles it — without wondering, afterward, exactly what it spent and why.
Businesses that accept agent purchases
You want to know a charge came from a real, human-approved authorization — not a hallucinated one — and to have a record that says so.
Teams building agent platforms
You need a trust story for letting your users' agents spend on a saved card. This is a story you can point to, adopt, and show.
Why this matters now
Each point stated only as far as the public record supports it.
Agents are starting to transact on saved cards, and the checkout rails are racing ahead — but a card handed to an autonomous agent has had no shared standard for staying inside a human's limits.
Agentic-commerce rails (OpenAI ACP, Google AP2, Visa, Mastercard), 2025–2026
The FTC's fake-reviews rule carries penalties up to $53,088 per violation, and Operation AI Comply has already acted against an AI company (Air AI, roughly $19M) over claims it could not substantiate.
U.S. FTC, 2026 (ftc.gov)
In May 2026 a German court issued a preliminary injunction over an AI assistant's output — an early signal that responsibility for what an AI does can land on the business behind it, not just the model.
Reported preliminary injunction, May 2026
When an agent overspends or buys the wrong thing, someone owns the fallout. A receipt proving the agent stayed inside a human-approved budget is the difference between an incident and an audit trail.
Design rationale
The buy-side twin of Nymrel Verified
Nymrel Verified publishes a machine-readable record of what is true about a business, so agents recommend it correctly. The Comfort Layer publishes a machine-readable record of what an agent actually bought, under what authorization, within what limits. Same verifiable Nymrel Record format — one is the sell side, this is the buy side.
Questions
Is this payment processing?
No. It sits on top of whatever payment rail you already use. It decides whether a charge is allowed against the human's policy and produces a receipt once it is — it never moves money itself.
Does it stop fraud?
It bounds and records what your own agent does. It is not a fraud, dispute, or chargeback system. What it gives you is agent spending that is limited, inspectable, and receipted.
Do I have to use Nymrel to adopt it?
No. The reference implementation is a zero-dependency TypeScript package you can drop into any agent flow, and the spec is open. Nymrel publishes the format and its own example; you own your receipts.
What is a Purchase Receipt?
A machine-readable record for a single charge — a Nymrel Record, the same verifiable format behind Nymrel Verified. It states what was charged and the authorization it charged against, and it honestly marks what it does not know at charge time, like whether the item was delivered.
What does it cost?
The spec and the reference implementation are open. Adoption is the point. Any future paid layer would be certification or a hosted audit trail, and would be named as such.
Building on agents that spend? Tell us what you're working on.