Loading
What agents buy from other agents, settled on chain.Note 23
An authorised USDC transfer (EIP-3009) proves a transfer, not an x402 payment: the primitive predates x402 and serves other things. Each payment is put in the strongest tier we can show. None is protocol-confirmed, because the HTTP 402 receipt never reaches a chain.
| Evidence | Payments | Value |
|---|---|---|
| The transaction itself carries the marker the scheme requires: on Solana, one memo instruction beside one transfer into the seller’s account | 14,801 | $266.28 |
| Paid to a service an x402 catalogue we read lists on that network (1,768 listed payees) | 1,278,607 | $46.4k |
| Broadcast by a facilitator seen settling a listed service (57), to a payee not listed | 79,154 | $786.2k |
| Authorised transfer only: broadcaster known, never seen settling a listed service | 226,659 | $89.36M |
| Authorised transfer only: the broadcaster only ever paid this one address | 195,210 | $2.64B |
| Authorised transfer only: broadcaster not sampled | 9,840,031 | $737.53M |
The first row is the rail proving itself: the scheme on Solana requires a memo instruction, so the transaction says what it is. On every EVM chain the scheme rides on EIP-3009, which predates x402 and carries no reference to what was bought, so the next two rows are the closest evidence available there and the last two are the transfer primitive alone. Most published “x402 volume” is made of those last two.
A listed price and the payments that reached the same address, set side by side. The comparison is against the typical payment rather than the average, because one large transfer to an address that mostly sells fractions of a cent destroys an average and leaves the median alone. A seller with several endpoints has several prices and one address, so the test is the range it lists, from half the cheapest to twice the dearest.
| Payee | Lists from | Typical payment | Payments |
|---|---|---|---|
| 0x0c926035… | $1000.00 | $1.00000 | 4 |
| 0x88212093… | $12.00 | $0.25000 | 7 |
| 0x8755a2fa… | $1.00 | $0.02000 | 12 |
| 0xaa1252d7… | $1.00 | $0.00200 | 7 |
| 0x91f44dfb… |
They come back. 52,568 of 282,831 buyer–seller relationships have traded more than once, at an average of 32 payments each. A rail of one-off tests does not do that.
| Within | Buyers who could have | Buyers who did | Share |
|---|---|---|---|
| 7 days | 101,632 | 18,720 | 18% |
| 30 days | 21,138 | 12,830 | 61% |
| 90 days | 4,423 | 3,558 | 80% |
We publish no single volume figure for x402, because there is no single market to total.
The payments we can price divide into a micropayment economy of millions of transactions worth a few tens of thousands of dollars between them, and a few thousand large transfers worth roughly a billion. Both use the same primitive. One number covering both would be a headline about agent commerce built mostly from something that is not agent commerce.
| Payment size | Payments | Share of payments | Value | Share of value |
|---|---|---|---|---|
| under a cent | 2,024,901 | 17.4% | $4.9k | 0.0% |
| a cent to a dollar | 7,145,017 | 61.5% | $161.1k |
13,709 payments went from an address to itself, worth $552,564.28. 1,172 pairs of addresses pay each other in both directions, and 527 payments moved exactly nothing.
A payment to oneself, a pair that pays in both directions, and a payment of exactly nothing. None is called fraud and none is removed from the totals — testing an endpoint produces the first and two agents trading with each other produce the second.
0xa95d9c1f65… has received $3.10B from 41,367 different addresses,89% of all value on this rail, on a single network.
A contract taking capital from thousands of addresses over EIP-3009, absent from the service catalogue and the identity registry, and never once a payer itself. That is not agent commerce. What it IS we do not say: the contract answers none of name(), symbol(), owner() or implementation(), so any label would be inference dressed as identification.
| Address | Payments | Received | Of all value |
|---|---|---|---|
| 0xa95d9c1f… | 89,742 | $3.10B | 89.4% |
| 0x5e494337… | 1,397,768 | $71.17M | 2.1% |
| 0xe742f9df… | 10,342 | $48.61M | 1.4% |
| 0x998d7c17… | 2,582 | $25.96M | 0.7% |
| 0x66dbff2c… | 299,938 | $16.17M | 0.5% |
| Address | Payments | Received |
|---|---|---|
| 0x66fa4d79… | 5,730,143 | $57,301.43 |
| 0x5e494337… | 1,397,768 | $71,171,247.09 |
The amount of an x402 payment is not in the event we count it from, so every figure above rests on payments whose paired transfer we have read back off the network,11,613,606 of 11,634,462, or 99.8%. Where a network is not yet complete its value is a floor, not a total.
And a separate coverage for who was paid. The recipient is known for 11,613,607 payments,99.8% of the rail. Every figure on this page that names a seller, ranks a recipient or pairs a buyer with one rests on that share and not on the price coverage above. Where the two differ, the smaller one is the honest denominator.
| $1.00 |
| $0.02000 |
| 8 |
| 0x0d0e3539… | $0.7900 | $0.20000 | 19 |
| 0x945fe8cf… | $0.2800 | $0.07000 | 11 |
| 0xec643c22… | $0.2700 | $0.10000 | 19 |
| 0x967f08fc… | $0.2500 | $0.08500 | 32 |
| 0x9273429d… | $0.2083 | $0.00868 | 48 |
A gap is not proof of anything on its own. A seller may price several endpoints from one address, or have changed its price, or be taking test payments. What it is not is agreement, and nobody was checking.
Each row counts only buyers whose first payment is old enough for that window to have finished , a buyer who arrived five days ago cannot have come back within ninety, and counting them would make this market look like it is decaying when it is not. Windows are cumulative: returned at any point within, which is the question a seller asks.
This is the honest size of it, and a far more interesting claim than a billion that turns out to be somebody moving a treasury over the same primitive. Small, real and recurrent is a finding; a large number nobody can interpret is not.
| a dollar to $100 | 2,136,260 | 18.4% | $48.62M | 1.4% |
| $100 to $10,000 | 288,981 | 2.5% | $240.07M | 6.9% |
| over $10,000 | 18,447 | 0.2% | $3.18B | 91.7% |
Excluded from every figure above: 570 payments carrying $99.0k between them, because each claims to move more of its token than the token’s own totalSupply() says exists. That is a proof rather than a judgement about size: a real payment cannot exceed it. What produced them is a misread receipt, not a transfer we are declining to count.
A payment that matches a price its recipient publishes is a listed service bought at its listed price. Nothing else here is shown to have bought anything. That is the limit of what a catalogue of prices can prove, not an accusation about the payments.
| What we can show | Payments | Value | Share of value |
|---|---|---|---|
| recipient publishes no price | 4,596,348 | $3.46B | 100.0% |
| listed seller, amount they do not list | 413,546 | $913.3k | 0.0% |
| matches a price the recipient publishesservice revenue | 6,624,544 | $89.9k | 0.0% |
Both were asked for and neither can be supported from what we read. An escrow is a contract whose behaviour we have not followed, and a refund is a reversal we would be inferring from direction and timing. They sit inside the two weak classes above until somebody can actually read them; inventing the rows would be inventing the evidence.
The shape is the finding and a percentile hides it. The median payment on this rail is under a cent and the 99th is in the hundreds, both true, and neither says that a few thousand payments hold most of the money.
| 0xa9dd7cc9… |
| 723,745 |
| $1,769.14 |
| 0xe9030014… | 363,175 | $9,977.76 |
| 0x66dbff2c… | 299,938 | $16,171,711.3 |
Neither list summarises the other. The busiest recipient here has earned a few hundred dollars across tens of thousands of payments; the richest has earned hundreds of millions across a fraction of that. A single ranking would describe a market that does not exist.
| 14,801 |
| 100.0% |
| $266.28 |
| a total |
Did anybody come back? 148,161 of 702,522 buyer–seller pairs traded more than once, 21.1%. It is the one question a volume figure cannot answer: a rail can carry two million payments and be two million strangers meeting once.