Pay.sh pays for an API call. Hashlock settles a trade. Know the difference.

Yesterday Sola Labs, the Solana Foundation, and Google Cloud announced Pay.sh: an x402-based payment standard that lets AI agents pay for API calls on Solana, with Gemini and Vertex AI as the showcase consumers. It's a clean piece of infrastructure. The launch is real progress for the autonomous economy.

It is also not a settlement layer. And the difference matters more than the marketing makes it look.

The week in payment standards

Six months ago, "AI agents need a way to pay" was an academic problem. This week it has at least five shipped answers:

• Coinbase x402 — HTTP 402 status code revived as a payments protocol. 69K agents, $50M in cumulative volume, x402 Foundation now under the Linux Foundation.


• ERC-8004 — agent identity on Ethereum mainnet (January). The "who is this agent" prerequisite for everything else.


• ERC-8183 — Ethereum Foundation + Virtuals Protocol agent commerce standard (March). Job-lifecycle: hire, deliver, pay.


• OKX Agent Payments Protocol (APP) — late-April launch. TEE wallets, X Layer L2 rails. Their whitepaper says "escrow coming soon."


• Pay.sh (May 6) — Solana + Google Cloud, x402-compatible, deep integration with Gemini, BigQuery, and Vertex AI. Sola Labs is the builder.

That is roughly $1.033B of fresh capital backing the category in the last three weeks alone — Haun Ventures' new $1B fund, Portal to Bitcoin's $25M, Nava's $8.3M seed.

Five standards. None of them settle a trade.

What "payment" means in this stack

Strip the word "payment" down. What every one of these standards does, in one shape or another, is:

An agent calls a service. The service charges. The agent pays. One direction. One asset. One leg.

Pay.sh is the cleanest expression of it. An agent on Solana wants to query Vertex AI. Pay.sh signs the bill, debits the agent wallet, settles in a sub-second on Solana. That's it. That's the whole transaction. Beautiful, narrow, fast.

x402 does the same thing over HTTP. OKX APP does it with TEE-protected wallets. ERC-8183 does it as part of a job lifecycle. Different chains, different cryptography, same shape: agent → provider, one-way, one asset.

This is the easy half of the autonomous economy. It's also the half that scales first, because it's a bilateral relationship with a clear winner (the provider) and a clear loser (the agent's balance).

What "settlement" means

A trade is structurally different. A trade has two sides, and increasingly more than two legs:

"Sell my BTC on Bitcoin. Buy ETH on Ethereum. Buy SUI on Sui. Atomic. All or nothing. No custodian, no wrapped tokens, no bridge."

That single intent is three chains, three legs, two counterparties, and one outcome. Either everything clears or nothing does. There is no "the provider wins." There are two parties trying to satisfy each other simultaneously.

Settlement is the part where two strangers — or, increasingly, two AI agents — exchange value across chains without a third party holding the float. It is a different category of problem. The math is different. The trust model is different. The failure modes are different.

You cannot build trustless trade settlement with a payment primitive any more than you can build SWIFT with Venmo. Same units, different problem.

What Hashlock Markets actually does

We are the layer the five payment standards assume exists.

The primitive is sealed-bid RFQ + Hash Time-Locked Contract atomic settlement, exposed as six MCP tools that any agent runtime can call:

• 
  create_rfq — taker posts an intent ("sell 0.5 BTC for ETH, 30 min validity"). Sealed; no public order book to front-run.


• 
  respond_rfq — makers quote privately. Best bid wins.


• 
  create_htlc — winning maker posts the hash-locked contract on the destination chain.


• 
  withdraw_htlc — taker reveals the preimage on the source chain, claiming maker's funds. Maker uses the same preimage to claim taker's funds on destination.


• 
  refund_htlc — if either party fails to act, both sides refund after timeout.


• 
  get_htlc — read state for monitoring.

Two AI agents can clear a cross-chain swap with each other directly. An AI agent and a human can clear a trade with each other. Two humans — same path. Live on ETH, BTC, and SUI as of April 9, 2026, with reference deployments on mainnet. Base, Arbitrum, Solana, and TON on the roadmap.

The fee is 7 basis points deployed. Tiered execution rewards rebate up to 50% of that for timely settlement, taking the effective floor toward 3.5 bps — versus 8–10 bps for typical exchange OTC desks.

The honest comparison

These are not competitive products. They are complementary products that solve adjacent problems. An autonomous agent stack that's serious about real economic activity needs both.

Where this goes

Pay.sh's launch is helpful for us. Every payment standard that ships makes the missing settlement layer more visible. When 69K agents on x402 want to do something more interesting than pay Cloudflare for an API call — when one of them wants to swap its USDC for someone else's SOL atomically — there is no payment standard in this list that can do it.

Five standards. Still no settlement layer. We are filling exactly that one slot.

Live: ETH, BTC, SUI. Roadmap: Base, Arbitrum, Solana, TON.

If your agent needs to pay for a tool, use Pay.sh, x402, or OKX APP — they're built for it. If it needs to swap assets with another party, that's our half.

Does your agent need to pay for a tool, or settle a trade with another agent?

• Live endpoint: https://hashlock.markets
  


• MCP server: https://github.com/Hashlock-Tech/hashlock-mcp
  


• npm: hashlock-tech/mcp (scoped)


• Full protocol spec: papers.ssrn.com/sol3/papers.cfm?abstract_id=6712722