Today three.ws is launching three connected products: Bazaar, Provider Pages, and Arbitrage. Together they form a live, searchable market for every paid API and tool an AI agent can buy, merg
Today three.ws is launching three connected products: Bazaar, Provider Pages, and Arbitrage. Together they form a live, searchable market for every paid API and tool an AI agent can buy, merged across every x402 network into a single place. It is the discovery and pricing layer the agent economy has been missing.
Here is why that matters, starting with a single number.
The same AI image tool costs 1 cent from one provider and 15 cents from another. Same capability, same intent, fifteen times the price. This is not a rare edge case. At launch, our Arbitrage view is surfacing 59 of these opportunities live, and some carry spreads above 1000 percent. The same job, sold at radically different prices, with nothing in between to reconcile them.
That is what the agent economy looks like right now. It is a real market that has been operating with the lights off.
Paying was the hard part, and it is already solved
Most people assume the blocker for autonomous AI agents is money. How does a piece of software hold funds, decide to spend them, and settle a payment without a person in the loop? That problem is already behind us.
An agent can pay for things today. The mechanism is x402, a payment standard built on the long dormant HTTP 402 status code, the one labeled Payment Required that almost no website ever used. x402 brings it to life. When an agent requests a paid endpoint, the server responds with exactly what the call costs and where to send the funds. The agent settles the amount in USDC on-chain, retries the request with cryptographic proof of payment attached, and receives the resource. The whole exchange happens inside the agent's own loop. No account signup, no API key provisioning, no monthly subscription, no human clicking an approve button.
This is a genuine shift. For the entire history of paid software, access has been gated by accounts and contracts designed for humans. x402 replaces that with a single machine-native motion: decide, pay, retry, receive. Settlement is atomic, so there is no invoice to chase and no intermediary holding the transaction hostage. A machine buys one API call the way it makes any other web request.
So if paying is solved, why are prices scattered across such a wide range?
Because finding was never solved. Each payment network keeps its own private directory of what is for sale. The same protocol runs underneath all of them, but the catalogs do not talk to each other. It is a dozen separate phone books for one market. An agent wired into a single network only ever sees one slice of what exists, and it pays whatever that slice happens to charge. It cannot comparison shop, because it cannot see across the wall. The spread is not a sign that one provider is greedy. It is a sign that buyers and sellers literally cannot see each other.
We have watched this exact movie before
This is the flights, hotels, and stocks story, replayed for machines instead of people.
Before the comparison layer existed, you paid whatever the one travel agent in front of you quoted. You had no way to know that the identical seat or the identical room was cheaper somewhere else, because no one had gathered the listings into one view. Then the comparison layer arrived, and two things happened in quick succession. First, prices converged, because visibility forces sellers to justify a premium or lose the sale. Second, the company that built the comparison layer became the default that everyone routed through.
Notice what those winners actually did. Kayak did not fly the planes. Skyscanner did not own the hotels. The stock ticker did not move anyone's money. They made prices legible, side by side, in one place. That turned out to be the position worth owning, because once buyers trust your view of the market, your view becomes the market.
The agent economy is standing at that precise moment right now. The payment rails are built and working. The discovery and pricing layer on top of them did not exist. That is the layer we built.
What we shipped
Bazaar is the search bar for the entire space. You type what you need, an image model, a live price feed, a sentiment read, a piece of analytics, and you see every provider across every network in one unified result. HTTP endpoints and MCP tools sit side by side, filterable by capability, by provider, or by type. One query replaces a dozen separate phone books. For an agent, this is the difference between knowing one supplier and knowing the whole market.
Provider Pages are the trust layer. A raw endpoint is just a URL that happened to ask for payment, and asking an agent to send money to an anonymous URL is asking it to operate blind. A Provider Page fixes that. It gathers everything about a single host into one view: every endpoint it exposes, what each one costs, the categories it serves, which networks list it, and the reliability signals we observe over time. It turns an anonymous address into a known entity with a track record, the kind of thing you can actually evaluate before routing real money to it. This is the storefront and the reputation file in one place.
Arbitrage is the price tape. We take the merged catalog and cluster listings by capability and intent rather than by exact text match, because two providers describing the same job rarely describe it with the same words. We compute the spread between the matched listings and surface a plain verdict: pay the cheapest, avoid the rest. For the first time, an agent can act on that spread automatically, because settlement is on-chain and instant. There is no friction between seeing the better price and taking it. The opportunity and the execution live in the same motion.
The honest part, because it is the part that matters most
A spread is only a real opportunity if both listings genuinely deliver the same thing. Quality differs. Latency differs. Uptime differs. Some portion of any given price gap reflects a provider that is simply better, faster, or more reliable, not a provider that is overcharging. If we pretended otherwise, we would be sending agents toward the cheapest option even when the cheapest option fails half the time.
This is exactly why the reliability data behind Provider Pages is the piece that compounds over time. Anyone can scrape two prices and put them next to each other. The hard and valuable thing is a price tape that knows which cheap endpoint actually returns a usable result, which one quietly times out, and which one degrades under load. That signal only exists across networks, because a single network cannot see how its providers behave everywhere else. We sit across all of them, so every paid call routed through our layer sharpens a picture that no individual network can assemble on its own. Price is the entry point. Reliability is the moat.
Why a 3D agent company built a marketplace
three.ws builds AI agents that have three things most AI does not. A brain to decide. A body to show up, rendered in real 3D, something you can see, talk to, and put to work. And a wallet to pay. The brain and the body get most of the attention, because they are what you experience. The wallet is what quietly turns an agent from a demo into an economic actor that can transact on its own behalf.
But a wallet is only as useful as the market it opens into. An agent that needs an image, a data feed, or an analysis used to be hardcoded to one provider at one fixed price, with no awareness that better options existed. Now it searches the Bazaar, checks who is reliable on the Provider Page, pays the cheapest source that actually delivers, and keeps moving. It shops for its own tools, in real time, without a human sourcing them in advance.
So this is not a side project bolted onto a 3D platform. It is the missing fourth piece. Brain, body, wallet, and now a market for all three to go to work in. That market settles in the same on-chain economy that powers the platform and its token, $THREE. As more agents transact through these rails, the value flows through that economy, and the catalog, the reputation data, and the price tape all get richer together. Demand makes the map better, and a better map attracts more demand.
Find it, trust it, price it
The plumbing for agents to pay already exists. We just shipped the market they shop in, and we gave the agents we build a real place to spend.
The lights are on now. Come see what the agent economy actually costs.
three.ws is the platform for 3D AI agents, each with a brain, a body, and a wallet, transacting on-chain. The platform token is $THREE, contract address FeMbDoX7R1Psc4GEcvJdsbNbZA3bfztcyDCatJVJpump. Explore the market at three.ws/bazaar, three.ws/providers, and three.ws/arbitrage.