AQUIX

The Eyes of the
AI Agent Economy

Independent observatory. Measuring which addresses actually earn on the x402 rail, from public blockchain evidence.

Explore the Registry  → How it works
Retained · all routes · lifetime
0.151%

Of every dollar that reached these addresses, that is what stayed.

service — 456 of 521circular flow — 31 of 521conduit — 21 of 521wallet — 13 of 521521published
x402 revenue, rail-verified$16,687,289.06
What these numbers mean
What these numbers mean

$564,467.78 kept of $374,754,349.42 received. Retained and processed are never summed — payment processors move money rather than earning it.

Kept is an upper bound — clamped at zero, USDC only. Received is exact. Both reconstructed to the cent from public Base state. Verify: this ratio · kept · received.

Verify: published addresses · settled · paid onward. 84 publicly named. No claim is made that any of them is an AI agent. The rail figure is 4.5% of all receipts.
73.8% of that rail figure is a single address — 3 payers, and it kept $95.78. We publish the concentration rather than remove the row: it passed the same rules as every other, and a total that rests on one payer set is something a reader is owed. It forwards 99.99948% of what it receives. It is still published as a service because our conduit test identifies a router or a rail and this address is recorded as a direct recipient — the same test, applied to all 521 published addresses, reclassifies none of the 14 that share this shape.
  • 456 service
  • 31 circular flow
  • 21 conduit
  • 13 wallet

Measured daily at 07:00 UTC · snapshot 2026-09-20 · volume processed $374.75M

Where the money went · all routes · lifetime
Received, all routes $374,754,349.42 Of which $16,687,289.06 (4.5%) settled through the x402 rail — exact, and the only part we can verify as x402.
Kept $564,467.78 What stayed — 0.15 cents of every dollar received. An upper bound: clamped at zero, USDC only.
Outflow $374,189,881.65 What left again. A lower bound, by the same clamping — 161 of 521 addresses read zero.

Kept and Outflow sum to the Received figure above, to the cent. The rail figure sits inside Received — it says how much of the money we can verify as x402, not where it went, and is never summed with the other two.

What we check · and how often we are wrong
33.2%

of what we publish rests on a single sampled payer — 173 of 521 addresses. Payer diversity is our own named signal, so these are the weakest verdicts we publish. They are not known to be wrong. Those 173 addresses hold $128,696.72 between them — 0.03% of everything received, a median of $44.50 each. The count is the honest headline; the money says where the doubt actually sits. Separately, 0 of 521 carry a verdict our stored evidence actually contradicts. That second figure is a lower bound and nought is not a claim of accuracy: it cannot see an agent we wrongly left out, nor a wrong verdict that is internally consistent. How it is measured · the raw measurement.

Reconciliation Report #1 — a second implementation, written and run by AQUIX, sharing no code with the pipeline. 948 checks, every disagreement adjudicated in the open, including the reconciler's own errors. Not a third-party audit.
Published 2026-07-23 · unlinked 2026-08-05 · restored 2026-09-12.

Check #01 — we re-implemented the seller rule Visa and Artemis published, on our own settlement store, and report where we disagree with them and where we think the disagreement is ours.

Corrections — every figure we have published and later found wrong, with what caused it and what it did not affect. Public and permanent; nothing here is deleted when it turns out to be wrong. A page has been unlinked once — the reconciliation report, and its dates are above.

The same number, in plain words

They kept about an eighth of one percent.

Of every dollar that reached these addresses, that is what stayed. Retained and processed are separate measures — payment processors move money rather than earning it. The blue sliver below is the entire verified retained revenue on this rail. The rest is money in transit.

Received, all routes · $374,754,349.42
$564,467.78 retained — 0.151% of everything received
Kept is an upper bound, clamped at zero. Received is exact. Both reconstructed to the cent from public Base state.
Daily volume · 2026-07-07 to 2026-09-18

A missing day is a hole, never a zero.

Every finalised day AQUIX has published, straight from the permalinks at /day/. On 3 mornings the observatory did not run — 2026-08-30, 2026-08-31 and 2026-09-02. A server storage failure stopped the 07:00 run on those mornings. No money was lost: the settlements were recovered from chain afterwards (correction #6, repaired again in correction #8). But a verdict is an observation made on a given morning, not a figure that can be recalculated later, so those days stay empty on purpose. We will not fill them with a guess.

0$457k$915k$1372k$1830k2026-07-07 observed2026-07-08 observed2026-07-09 observed2026-07-10 observed2026-07-11 observed2026-07-12 observed2026-07-13 observed2026-07-14 observed$1,239k2026-07-15 observed2026-07-16 observed2026-07-17 observed2026-07-18 observed2026-07-19 observed2026-07-20 observed2026-07-21 observed2026-07-22 observed2026-07-23 observed2026-07-24 observed2026-07-25 observed2026-07-26 observed2026-07-27 observed2026-07-28 observed2026-07-29 observed2026-07-30 observed2026-07-31 observed2026-08-01 observed$1,010k2026-08-02 observed2026-08-03 observed2026-08-04 observed2026-08-05 observed2026-08-06 observed2026-08-07 observed2026-08-08 observed2026-08-09 observed2026-08-10 observed2026-08-11 observed2026-08-12 observed2026-08-13 observed2026-08-14 observed2026-08-15 observed2026-08-16 observed2026-08-17 observed2026-08-18 observed2026-08-19 observed2026-08-20 recovered from chain — no verdict was taken that morning2026-08-21 recovered from chain — no verdict was taken that morning$955k2026-08-22 recovered from chain — no verdict was taken that morning2026-08-23 recovered from chain — no verdict was taken that morning2026-08-24 recovered from chain — no verdict was taken that morning2026-08-25 recovered from chain — no verdict was taken that morning2026-08-26 recovered from chain — no verdict was taken that morning2026-08-27 recovered from chain — no verdict was taken that morning2026-08-28 recovered from chain — no verdict was taken that morning2026-08-29 recovered from chain — no verdict was taken that morning2026-08-30 not observed, no snapshot was taken that morning2026-08-31 not observed, no snapshot was taken that morning2026-09-01 observed$1,262k2026-09-02 not observed, no snapshot was taken that morning2026-09-03 observed2026-09-04 observed2026-09-05 observed2026-09-06 observed2026-09-07 observed2026-09-08 observed$928k2026-09-09 observed$1,688k2026-09-10 observed$1,372k2026-09-11 observed2026-09-12 observed2026-09-13 observed2026-09-14 observed2026-09-15 observed2026-09-16 observed$1,777k2026-09-17 observed2026-09-18 observed$1,248k07-1508-0108-1509-0109-15not observed2 of 3 · also 09-02USDC processed per day · paler outlined bars were recovered from chain after the fact — the money is measured, the verdict is not · hatched days were never observed and are left empty, not zeroed

Swipe the chart sideways to read the whole period →

Growth · measured day against measured day

The rail is growing, and here is the arithmetic.

What this measures
What this measures

Basis: every x402 settlement on Base is swept, not sampled. The collector enumerates every EIP‑3009 authorisation on the chain and does not start from a list, so a payment we have never seen before is still counted — 1,280 addresses settled money here, not only the 521 published in the register. The register is the narrower subset that also passed the classifier, the sybil gate and the scope test; published agents account for $16,687,289.06 of the $39,095,224.62 above — the rail-verified figure at the top of this page. What this is not: all of Base. The chain carries far more than x402, and an address whose inbound is not x402 — a router, a bridge — is excluded here even when it is very large. 71 of 74 calendar days were measured; the rest are excluded rather than counted as zero.

Every figure below is an average per measured day, never a window total. A window total counts a morning we did not run as zero; an average simply has one fewer sample. The 3 mornings we did not measure are excluded, not zeroed.

Settled volume per measured day
+79.5%
$395,862.20 → $710,581.19
first 35 measured days against the most recent 35
Addresses paid per measured day
+149.4%
172 → 430 addresses
who was paid, not how much · peak on the record 678 · counts every address paid that day, including those later judged bot-shaped or empty
Last 7 measured days
-14.9%
$960,566.83 → $816,982.64
against the 7 measured days before them · a short sample, and it shows
Measured in total
$39.10M
across 71 measured days
2026-07-07 to 2026-09-18
Recency

How much of the register is alive

A high row is not necessarily an active one.

  • Paid within the last 7 finalised days253  48.6%
  • Paid, but longer ago than that167  32.1%
  • No finalised x402 day on record at all101  19.4%

521 published agents · figures trail today by two days

Confidence

How sure we are

Our own uncertainty, published rather than hidden.

521AGENTS
High9718.6%
Medium17734.0%
Low24747.4%
The denominator

What is actually on this rail

Every address paid on x402 in the 7 fully-swept days to 2026-09-19, sorted by what the money did. No claim is made that any of them is an AI agent — that cannot be proven from chain data, so we report economic behaviour instead. Download the raw census

CONDUIT — $4,013,459.96 (43.4%)INFRASTRUCTURE — $1,947,723.03 (21.1%)UNDECIDABLE — $1,822,329.28 (19.7%)PASSENGER — $1,090,462.81 (11.8%)EARNER — $365,220.46 (4.0%)BOT-SHAPED — $2,988.85 (0.0%)$9.2Mreceived
  • CONDUIT43.4%
  • INFRASTRUCTURE21.1%
  • UNDECIDABLE19.7%
  • PASSENGER11.8%
  • EARNER4.0%
  • BOT-SHAPED0.0%

23,466 addresses · $9,242,184.39 received in the window. AQUIX publishes 521 of them — the ones that pass the classifier, the sybil gate and the scope test. Everything else is shown here so the denominator is visible.

Ranked by retained, not by volume

Who actually kept the money

The registry ranks by revenue we can verify. This is the other ranking — by what was retained, the number AQUIX exists to report. Where an address sent most of its money back to the payers who sent it, the row says so — the figure is what remained, not what it earned.

USDC retained, lifetime · these 8 hold $479,303.18 of the $564,467.78 the register kept (84.9%) · 2 of these have no x402 settlement on record (101 in the register); the kept figure is reconstructed from Base state, not corroborated by a settlement

Independent

No protocol can pay to change a verdict. AQUIX takes nothing from the agents it measures.

Verifiable

Every figure is reconstructed from public Base state. Each agent page shows the address, so you can check it yourself.

Transparent

The method is published and every correction is permanent. Nothing is rewritten after the fact — and when a page has been unpublished, the dates are printed beside it.

Measured daily

One snapshot every day at 07:00 UTC, timestamped and archived. When a morning is missed, the record says so.

Evidence before trust

Verdicts follow published rules, not opinion. Where a number does not exist, the page says so.

Verified retained revenue $564,467.78 · volume processed $374,754,349.42 across 521 published addresses · snapshot 2026-09-20 · measured daily at 07:00 UTC, published when reviewed

AQUIX Evidence before trust @aquixai intelligence@aquix.ai
The record
All published agentsKnown, not measuredMethodology and rulesDefinitions
This page
Latest daily recordThe raw census (JSON)Sitemap
Why you do not have to trust us

Every figure is reconstructed from public Base chain state. No API key, no account, no trust required — open an address and compare the USDC transfers against what we show. Every verdict is adjudicated in public and logged with its date and reason; nothing is edited silently and no verdict is removed from the record.

AQUIX · an independent observatory of the x402 rail on Base · Measured daily at 07:00 UTC · Snapshot 2026-09-20 · Methodology v3.2 · Classifier v4 · Latest correction #9 · No protocol pays to appear here

Company names and logos are the property of their owners and are shown for identification only. Their appearance here does not imply any affiliation with, or endorsement by, AQUIX.