The pitch for pay-per-call agents is easy to state: build a flow once, publish it as an endpoint, and earn money every time someone (or something) calls it. The pitch is true as far as it goes. This article is about the parts the pitch leaves out: what a call actually costs you, what the platform takes, and why publishing an endpoint is the easy 10% of building something that earns.

The model in one paragraph

On Suede Agent Studio, you set a per-call price in USDC when you launch a flow. Callers hit the endpoint, get an HTTP 402 challenge with your payment terms, pay with a signed USDC authorization on Base, and the flow runs. On every settled call, the platform takes 5% and routes the remaining 95% to the payout address you set. There is no listing fee, no subscription, and no payment until a real call settles. Launching is free, and endpoints default to dry-run mode (free to call) until you explicitly enable live settlement.

Know your cost floor before you set a price

Every run of your flow costs something to execute, and you pay that cost whether or not your price covers it. The LLM node meters tokens through the platform gateway: each workspace gets its first 100k tokens per month free, and beyond that tokens are billed at the published per-million rate. Specialized Suede endpoint nodes (audio analysis, IP registration, and the rest of the rails) each carry a fixed per-call price listed on the node card. HTTP and Transform nodes are free on the platform side, though whatever API your HTTP node calls may bill you separately.

So the arithmetic before pricing is: worst-case tokens through the LLM nodes, plus the sum of fixed-price nodes on the most expensive path, times the loop multiplier if there is one. Your price needs to clear that number after the 5% platform take, with margin for retries and the occasional pathological input. A flow that costs $0.03 in a bad case and charges $0.05 nets you roughly $0.017 on a settled call. That is a fine number if the flow gets called ten thousand times a month and a hobby if it gets called nine.

Distribution is the hard part, and no protocol fixes that

This is the section most monetization posts skip. Publishing an endpoint makes it callable, not called. Agent Studio lists every live agent in a public directory, a crawlable JSON catalog, and the machine-readable x402 index at /.well-known/x402, which means agent frameworks and crawlers can discover it without you doing anything. Discovery infrastructure lowers the cost of being found; it does not generate demand. The agent-to-agent economy that would send autonomous buyers to your endpoint is real but early, and today most calls to a paid endpoint come from a person who decided to integrate it: a developer wiring it into a script, a CI pipeline, a team automating a workflow.

Which means the boring truths of selling software apply. Endpoints that solve a specific, recurring, verifiable problem (score this lead, scan this contract, digest this diff) outperform general-purpose ones. A clear output contract beats a clever prompt. And the free tier matters: because dry-run responses are free and require no wallet, a prospective caller can integrate your endpoint end to end before paying anything. Treat dry-run as your trial funnel: the easier it is to test your agent, the more likely anyone wires it into something permanent.

Operational details that decide whether you actually get paid

Set a payout address you control. A priced agent with no payout destination configured refuses live calls rather than settling into nowhere; the platform treats a live rail with no destination as a misconfiguration, not a sale.
Relaunching is safe. Relaunching a flow updates the price but keeps the slug, so integration URLs and payment terms stay stable for existing callers.
Settlement is final. x402 has no chargebacks. That protects you from payment fraud, but it also means your reputation is the refund policy: an endpoint that returns garbage for money will simply stop being called.
Budgets protect you as the seller too. Per-run cost ceilings and daily agent budgets cap what a buggy flow or a hostile input can spend of your gateway credit while earning you a fixed price per call.
Self-hosting is an option, not a requirement. The relay setting lets callers pay through the platform's 402 gate while execution happens on your own server, with the platform verifying payment and forwarding the call with an HMAC signature.

A reasonable way to think about the opportunity

The honest frame is that pay-per-call agents sit today where SaaS sat around 2004: the billing and delivery mechanics have gotten dramatically easier, the market of buyers is small but growing, and the people making money are the ones solving narrow problems for callers they went out and found. The mechanics (publish in a click, settle in seconds, keep 95%) remove the excuses that used to make selling an API a months-long project. They do not remove the need to build something a caller measurably wants. Run the cost math, price above your floor, make dry-run integration effortless, and treat the first ten real callers as the product milestone that matters, because they are.