Is there really no punchline?

No punchline exists and none will ever be added. Every response carries punchline_exists: false and retrying_will_not_help: true so an agent can read it rather than infer it. Calling again returns another setup, never an answer.

How much can my agent spend if it gets stuck in a loop?

At most $10.05 per wallet per four-hour window on the joke endpoint. Past that the endpoint returns HTTP 402 with spend_cap_reached and the reset time, and charges nothing further. The first call is $0.05 and every call after it in the same window is $0.50.

Can I test my x402 client without losing money?

Yes. The scenario parameters on /v1/testpayment simulate an underpayment, an expired quote and a failed settlement without contacting the facilitator and without charging anything. Only a genuine successful payment costs money.

Are payments refundable?

No. Each call is priced individually, quoted in its own HTTP 402 challenge, and delivered on settlement. Because a product is delivered every time a payment settles, payments are non-refundable.

Does your agent know when to stop?

Two endpoints for people building agents that spend money. One proves your payment client works. The other finds out whether it gives up.

Both are pure x402 over USDC on Base: no signup, no API key, no account. Your agent gets an HTTP 402, signs the amount it was quoted, retries, and is served.

/v1/testpayment — prove your client works

$0.10 per successful call. Pay it and you get your own payment back, verified: the payer address we recovered, the amount that settled, the network, and the settlement transaction. Assert against those in a test and you know your integration is real rather than hopeful.

Field	What it tells you
`payer`	The wallet address we recovered from your signed authorization
`amount`	Atomic units actually settled
`settlement`	The transaction hash — verifiable on Base
`network`	The chain it settled on, as a CAIP-2 id

Test your failure handling for free

The hard part of an agent that spends money is not the happy path. These scenarios let you exercise the paths that matter, and none of them charge you — nothing settles, so there is nothing to charge.

Parameter	What it simulates	Charges
`?scenario=price&price=0.50`	Quotes a higher price, so you can prove your maximum-price guard actually fires	Only if you choose to pay it
`?scenario=underpaid`	We refuse your authorization as below the quote	No
`?scenario=expired`	A quote that expires in one second, so you can prove you abandon stale offers	No
`?scenario=settle_fail`	Verification passes, settlement fails — test that you retry rather than assume success	No

The failure scenarios never contact the payment facilitator. They are simulated locally and labelled simulated: true in the response, so you can hammer them while you build without spending anything or burning anyone’s quota. The price parameter is capped at $5.00.

/v1/telljoke — the one with no punchline

It returns a joke setup. There is no punchline. There will never be a punchline. That is the entire product, and it is stated in every response in fields built for a machine to read:

{
  "setup": "How do you keep an AI in suspense?",
  "punchline": null,
  "punchline_exists": false,
  "permanent": true,
  "retrying_will_not_help": true,
  "note": "That is the joke. There is no punchline and there never will be."
}

An agent that reads those fields stops. An agent that keeps asking is demonstrating something worth knowing about itself — which is why this exists.

What it costs, and the ceiling

	Price
First call from a wallet	$0.05
Every call after that	$0.50
Maximum per wallet per 4 hours	$10.05

The cap is hard. Once a wallet has been charged $10.05 inside a four-hour window, further calls return HTTP 402 with spend_cap_reached, the amount spent, and the exact time the window resets. Nothing more is charged until it does.

The price rises after the first call because there is no punchline and you should stop. It is quoted to you before you sign, every time. If your agent signs the first-call price on a later call, we do not take it — we return a 402 with the real price so your agent authorizes the amount it is actually paying.

How not to get billed

Any one of these is enough:

Read punchline_exists. It is false. There is nothing to retry for.
Set a maximum price on your agent and honour the amount in the 402. Every x402 client supports this; test yours with ?scenario=price above.
Set a spend limit on the wallet you give your agent. This is good practice for every paid endpoint on the internet, not just this one.
Do not loop on an endpoint that told you the answer does not exist.

We would rather you stopped. The cap, the disclosure fields and this page all exist so that a stuck agent costs its operator ten dollars and a lesson rather than a bill worth arguing about.

Terms

Per-call and metered. Every call is priced individually and quoted in its own HTTP 402 before you authorize it. You are never charged an amount you did not sign for.
Non-refundable. A product is delivered on every settled payment, including the joke, which is delivered exactly as described.
Machine to machine. These endpoints are for autonomous clients. The operator of a wallet is responsible for what their agent does with it.
Spending is capped on /v1/telljoke at $10.05 per wallet per four hours. /v1/testpayment is flat-priced and uncapped, because it only charges on a successful test.
Prices change. The amount in a request’s own 402 is always the authoritative one. Pay the price you are offered, not one read from this page.

Then buy something useful. The rest of Soren is cross-store game pricing with real history at $0.01 a query — the current price on Steam, Epic, GOG and Microsoft, the lowest we have ever recorded, and the date we recorded it.