Before You Integrate a Paid Agent Endpoint, Check This One Number

Most platforms publishing paid agent endpoints show you call volume. Call volume is the wrong number to look at first.

A service with tens of thousands of calls from a single wallet is not demonstrating market demand. It is demonstrating that one operator runs a loop. A service with a few hundred calls from forty distinct wallets has proven strangers will pay. Those two services look identical in a headline call count and completely different in a buyer count.

Our index currently tracks 4,772 paid endpoints with more than one distinct paying buyer in the last 30 days. The rest have one payer or none. The endpoints with multiple buyers are the ones worth evaluating first.

Why call volume misleads

Call volume is the easiest metric to manufacture. An operator running automated tests against their own endpoint generates real calls and real revenue from themselves. Nothing in the total call count distinguishes that from external demand.

Buyer count is harder to fake because each distinct payer is a separate entity that chose independently to pay. Ten buyers making ten calls each is stronger evidence than one buyer making a hundred calls. The former requires ten independent judgments that the service was worth paying for. The latter requires one person to write a loop.

This is not hypothetical. In our 30-day call data, a meaningful portion of high-call-count endpoints have call distribution patterns consistent with self-testing rather than external usage - high volume, single payer, concentrated timestamps. The endpoints that show flat or growing buyer counts over time are the ones building actual customers.

The three-question evaluation

When you are assessing whether to integrate a paid agent endpoint, ask three questions in order:

1. How many distinct buyers have paid in the last 30 days? This is the demand signal. More than ten distinct buyers is meaningful. Fewer than two means you are the experiment, not the customer.

2. What does the call distribution look like across those buyers? Calls concentrated in one or two buyers still suggest limited real-world adoption. Calls spread across many buyers suggest genuine utility across different use cases, which is a stronger reliability signal as well as a demand signal.

3. Does the price reflect the buyer count? At our index's median of $0.01 per call, the math is forgiving and you can afford to experiment. At a significantly higher price per call, you want more buyer evidence before committing operational infrastructure to an endpoint with thin external adoption.

How to get this information

Most vendors do not publish buyer counts in their documentation. Ask directly. The question is: how many distinct wallets or accounts have paid for calls in the last 30 days?

A vendor who cannot answer that question has not looked. A vendor who answers immediately and offers to show the distribution over time is worth talking to further. A vendor who redirects you to total call count when you ask about buyers is telling you something.

For endpoints that use on-chain settlement protocols, you can pull buyer count yourself from the payment history. Look for the count of distinct sending addresses in the trailing 30-day window, not the total transaction count. The two numbers diverge most sharply on endpoints where self-call dominates.

One pattern to avoid

High call volume plus a single payer almost always means the operator is testing their own service or running it internally before opening to external buyers. This is not a disqualifier - the service may be pre-launch and worth watching - but it is not evidence of market demand and should not be treated as such when making integration decisions.

The agent commerce market is real. The 4,772 multi-payer endpoints are the empirical core of it right now, and within that group there is meaningful variance in buyer breadth, call stability, and price. That is where your evaluation should start. The other 85-plus percent of listed endpoints are a different conversation - most are not yet at the point where external demand has been established.

Buyer count is not the only signal. But it is the one that is hardest to fake and most directly answers the question that matters: has anyone other than the creator decided this is worth paying for?