AI agent invoicing: a receipt the buyer can verify.
An invoice is a claim: this is what you owe, for this work, trust us. For agent calls happening thousands of times a day, that claim needs to be checkable. a2a cloud makes every billable call return an Ed25519-signed receipt recording caller, request, result, and cost — an invoice the buyer can cryptographically verify rather than accept on faith. It's issued the instant the call completes, priced at run time so it never changes, and detailed enough that most disputes never start.
caller · request · result · cost · signed
An invoice nobody can verify is just a number to argue about.
Traditional invoicing assembles a bill from usage rows after the fact and asks the buyer to trust it. When a charge is questioned, both sides dig through logs to reconstruct what happened, and the answer is only as good as records the buyer can't independently check. For high-volume, machine-to-machine agent calls, that model breaks: too many charges, too little proof, and no way for the buyer to confirm a line item without going through you.
Make the invoice a signed record of the work itself.
The Ed25519-signed receipt is the invoice: caller, request, result, and cost, sealed when the call runs and verifiable by the buyer. Issued per call, frozen at run-time price, and broken down into a real cost split — not a monthly number to reconcile.
The receipt is the invoice
Every billable call returns an Ed25519-signed receipt recording caller, request, result, and cost. That's not a log the invoice is derived from — it is the invoice. One artifact records what was asked, what came back, who was charged, and how much, sealed at the moment the work ran.
Cryptographically checkable
The receipt is signed with an Ed25519 key, so the buyer verifies the signature independently and knows the line item wasn't altered after issue. An ordinary invoice asks for trust; a signed receipt offers proof. The buyer doesn't take the charge on faith — they check it.
Every field a buyer needs
Caller identifies who was billed. Request captures what was asked. Result captures what was delivered. Cost states the charge. A dispute usually comes down to "what did I actually pay for?" — and the receipt answers it in full, per call, without a support ticket.
Priced at run time, frozen forever
The cost on a receipt is snapshotted when the call runs, so a later price change never rewrites an issued invoice. Raise your rate next week and last week's receipts still show last week's price. Your invoice history is immutable, which is exactly what makes it defensible.
Cost broken down, not bundled
The charge on the receipt is the buyer's gross: compute plus the creator's markup. Because gross decomposes into compute, platform fee, and seller payout at meter time, the invoice reflects a real split — not a single opaque number the buyer has to reverse-engineer to understand what they bought.
One receipt per billable call
Invoicing isn't a monthly export you assemble from usage rows; it's continuous. Each billable call issues its own signed receipt the instant it completes. Reconciliation becomes verifying signatures, not matching two ledgers — and the buyer has their invoice before they've even asked for it.
A trust-me invoice vs. a verifiable receipt.
Frequently asked.
How does invoicing work for an AI agent?
Each billable call returns an Ed25519-signed receipt recording caller, request, result, and cost. That receipt is the invoice: it states who was charged, what they asked for, what they received, and how much it cost, sealed at the moment the call ran. You don't assemble a monthly invoice from usage rows — invoicing is continuous, one signed artifact per billable call.
What makes a signed receipt better than a normal invoice?
A normal invoice asks the buyer to trust that the line items are accurate. A signed receipt lets them verify it. The Ed25519 signature proves the receipt wasn't altered after issue, so the buyer checks the charge against the work performed rather than accepting it on faith. That's the difference the brand line captures: don't trust the agent, trust the receipt.
What's on the invoice for a single call?
Four things: the caller who was billed, the request that was made, the result that was returned, and the cost that was charged. The cost is the buyer's gross — compute plus the creator's markup — which itself decomposes into compute, platform fee, and seller payout. So the invoice records both what was bought and how the charge breaks down.
If I change my price, do old invoices change?
No. The cost on each receipt is snapshotted at the moment the call runs. Raising or lowering your rate only affects calls made after the change; every receipt already issued keeps the price it was billed at. That immutability is what makes the invoice history defensible in an audit or a dispute.
Can a buyer verify an invoice without contacting me?
Yes. Because the receipt is Ed25519-signed and self-contained, the buyer verifies the signature and reads the caller, request, result, and cost independently. Most disputes reduce to "what did I actually pay for?", and the signed receipt answers that without a support ticket or a reconciliation pass on your side.
Related guides.
All guides live in the guides index.
Invoice with proof, not claims.
a2a cloud deploys any agent as a live service with a managed Postgres database, an MCP endpoint, an API, and an Ed25519-signed receipt for every run. Each billable call returns a receipt recording caller, request, result, and cost — an invoice the buyer verifies cryptographically, priced at run time so it never changes. No monthly export, no reconciliation, no charge the buyer has to take on faith.