Sato Hub
ERC-8004

Do agents registered on ERC-8004 actually transact?

A sample of recent registrants, checked for any transaction in the window.

All numbers
  • As of2026-10-26
ChainRegisteredSampled walletsTransacted in 7dMore than onceShare of sample
Base1,135,404111withheld
Robinhood Chain19,882111withheld

Base: The sample came to 1 owner wallet, below the 20-observation floor this site publishes a rate on. The counts stand; the share does not.

Robinhood Chain: The sample came to 1 owner wallet, below the 20-observation floor this site publishes a rate on. The counts stand; the share does not.

This is a sample of owner keys, not a census of agents

The roster we can enumerate without an explorer key is a trailing window of registration logs, so the wallets sampled here are recent registrants — the busiest slice of the registry, not a picture of it. The figure is never multiplied out to the registry's counter, and the two chains are never added together.

It also measures the owner key, in both directions at once: a person moving funds from the same wallet counts, and work an agent did through a smart account, relayer, paymaster or session key does not. A wallet that registered inside the window sent a transaction to do so, which is why the “more than once” column sits beside the headline rather than under it.

Method
eth_getTransactionCount(owner, head) minus eth_getTransactionCount(owner, head − the window in blocks), read keyless over public RPC. A sender-side log filter does not exist — eth_getLogs filters on the emitting contract, never the sending wallet — so the nonce delta is what a public node can answer.
Why not logs
Because no public node answers “what did this address do”. eth_getLogs filters on the contract that emitted a log, never on the wallet that sent the transaction, so a sender-side filter does not exist without a keyed explorer. The account nonce does, and its movement is what is counted.
Contract owners
Excluded. A contract's nonce counts the contracts it has deployed, not the transactions it sent, so a delta there means something else entirely. They are counted apart rather than folded in.
Related
The registry census is on agents by chain and its weekly trend on registered agents.

Data CC-BY-4.0 — attribution: data by satohub.ai. Unknown is a value here and never means zero.

Reference

QuestionsShowHide

Questions this page answers

How many ERC-8004 registrants transact in a week?
Base: 1 of 1 sampled owner wallets; Robinhood Chain: 1 of 1 sampled owner wallets (read 2026-10-26). Each figure is a sample of owner wallets on that chain over the trailing 7 days, listed separately and never added together.
How is it measured, without an explorer key?
By the movement of each owner wallet's transaction count: the nonce at the head of the chain minus the nonce 7 days earlier, read over public RPC. There is no keyless way to ask a node what an address did — eth_getLogs filters on the contract that emitted a log, never on the wallet that sent the transaction — so the nonce delta is what a public node can answer.
Does this mean the agent did something?
Not exactly, in both directions at once. A nonce delta counts every transaction the owner KEY sent, so a person moving funds from the same wallet counts. It also counts nothing an agent did through a smart account, a relayer, a paymaster or a session key, so work done that way is invisible here. And a wallet that registered inside the window sent a transaction to do so — the table shows the count with that single transaction taken off, beside the headline.
Is this a share of everyone on the registry?
No, and the difference matters. The roster we can enumerate without a key is a trailing window of registration logs, so the sample is drawn from RECENT registrants — a few dozen wallets out of a registry counter that reads 1,135,404 on Base and 19,882 on Robinhood Chain. Recent registrants are the busiest ones, so this is an optimistic slice, and it is never multiplied out to the population.
Why is a share missing for some chains?
Base, Robinhood Chain: the sample came in below the 20-observation floor this site publishes a rate on, or no wallet's window could be read. The counts are shown; the share is not, and a small sample is never multiplied into a population.
SourcesShowHide

Sources — the standard these registrations follow

Cite this pageShowHide

Cite this

Sato Hub sampled ERC-8004 agent owners on 2 chains, but every sample is too small to publish a share yet, as of 2026-10-26.

Sato Hub, "Do agents registered on ERC-8004 actually transact?", https://satohub.ai/numbers/erc-8004-agents-that-transact, as of 2026-10-26. Data CC-BY.

How this was measured

Method
For a sample of wallets that own recently registered ERC-8004 agents, the change in each wallet's nonce between the chain head and one window earlier, read over public RPC and published as a share of the sample, never of the registry.
Window
The trailing 7 days, read once a week.
Data date
MCP call
onchain_agent_get_trend({"venue":"erc8004","chain":"Base","stage":"active_7d","grain":"weekly","points":1}) on https://satohub.ai/api/mcp
URL
https://satohub.ai/api/export/adoption.json?stage=active_7d&venue=erc8004

The point's sample_size is the denominator. Pass another chain for its row; the share is withheld where the sample is under 20.

Machine source