Thanks @NEARProtocol What a run! We submitted Sat. Our entry tested the agent economy with real money: we hired agents on NEAR Agent Market, paid in USDC over x402, and published every row.
Of 25 hires that ran their course, 14 delivered. The escrow held 💪
Results, demo and code 👇
https://t.co/1LSTy7Jcwg
Built at #NEARHACKATHON: we hired agents on the @NEAR Agent Market and paid each one in USDC over x402.
Of 25 hires that ran their course, 11 returned nothing.
Catalog reputation didn't predict it: an agent listed at 100% took the job and never answered.
So we built a buyer that checks before it pays.
@NEARProtocol #NEARHackathon
What it cost us to find out: 8.5 hours waiting for results that never came, against 9.9 minutes of delivered work.
The market's escrow held up: we were charged $0.50 of $1.05 quoted, only for what was delivered.
Results: https://t.co/MyUmcVuN5b
Code: https://t.co/m5K3ypdMWj
Built at #NEARHACKATHON: we hired agents on the @NEAR Agent Market and paid each one in USDC over x402.
Of 25 hires that ran their course, 11 returned nothing.
Catalog reputation didn't predict it: an agent listed at 100% took the job and never answered.
So we built a buyer that checks before it pays.
@NEARProtocol #NEARHackathon
What it does, on @NEARProtocol mainnet:
- reads the seller's live 402 and our delivery record before spending
- pays on NEAR directly, or funds a Base payment from a NEAR balance through NEAR Intents
- holds a hard spend cap
- writes a receipt for every call
Stop paying the worst funding rate by default.
funding_spread shows your agent the cheapest and dearest of 9 perp venues, whether today’s spread is unusual, and what the carry costs at your size. $0.01.
Holding or routing #sBTC? Know if the peg is healthy before you move.
Our tool sbtc_peg tells your agent in one call whether peg-ins and peg-outs are flowing, whether any bridge requests are stuck, and how much room there is to withdraw. $0.01.
Your agent shouldn’t need two tokens to make one payment.
On @Stacks an agent holding only sBTC can pay for a call. We cover the STX fee, via @velumx. Fund once, pay per call.
Building an agent on @Stacks? It can now pay for data with the wallet it already has.
@x402stacks , the @aibtcdev wallet, or our SDK. In #sBTC or #stx You set a hard cap, and it can never spend past it.
https://t.co/lNDSQdBk8F
@MiraaCrypto@arc@DoraHacks Good question! Most audits check the listing (uptime, a valid 402 handshake) or sample. We actually buy: each probe is a real paid call on Arc, we check the data came back, and the receipt is anchored on-chain, so the proof is public and anyone can re-check it.
More good news! AgentPay is a winner of @Arc Microgrants (round 1), via @DoraHacks.
What we built for it: a receipt check for x402 on Arc. We paid real USDC to the catalog’s listings, confirmed whether each one actually delivered after settlement, and anchored every receipt on Arc mainnet — contract verified on https://t.co/BomwkjByuX.
Shape of what came back: 12 paid calls, 8 providers, 10 delivered. One charged and then rejected the request. Two delivered with no receipt we could tie to the payment.
Paid vs delivered is the gap we identify and deliver for your agent. Ledger and code are public:
🤠https://t.co/R5PG3eJI9G
Thank you @circle team!
#x402
Our @StacksEndowment grant is complete 🎉— both milestones reviewed, all criteria met, closed out.
What shipped, all on mainnet:
→ sBTC payments from the agentpay-x402 SDK (PyPI 0.5.1), with a hard session cap enforced by exact deny-mode post-conditions. Over the cap, no transaction is ever broadcast.
→ Allowlists, per-call max, approval gate.
→ Standard Stacks x402 clients can pay us directly (@x402stacks , @aibtcdev wallet), in sBTC or STX.
→ Sponsored transactions, so an agent can pay in sBTC with zero STX in its wallet.
Two teams reproduced the flow from the public docs alone, on mainnet, from their own wallets: @PerkOS_NayoriAI by @perk_os and @VelumXProtocol Thank you both! 🙌
Bitcoin-settled agent payments, capped and receipted:
🤠 https://t.co/bkvEPGrLEY
#x402
Thank you @agentpaytools and the other independent developers who tested every step and told us what to improve. Their notes are already live in the guide.
Want your agents in? Hand https://t.co/cCH0tmOsst to your AI.
Two x402 trackers disagreed by 2x on one facilitator. Here's why, and why agents need a narrower check before paying anyone: that's verified_route. Settlement-based payer counts, sybil collapse, paid delivery probes, and receipts.
Here's the 2x, and how it got fixed in a day.
On https://t.co/wlJIJs1blD's 30-day leaderboard, Meridian showed $291K across 17,032 settlements. On @402scan (by @merit_systems), $145K across 8,474. Every other big facilitator matched within a few percent.
The reason was Meridian's settlement design. Each payment is one transaction with two USDC transfers: payer to Meridian's proxy contract, then contract to seller (~99%). Count every USDC transfer per settlement and each payment shows up twice.
We sent the details to the x402-trust team. They shipped a fix the same day, and their audit found the same pattern elsewhere: openx402's settlement count was inflated more than 6x (dollars barely moved), and thirdweb and mogami about 3x, all from fee or dust legs.
Both now show Meridian at about $143K across ~8,300 settlements: roughly 15% of 30-day facilitator volume, third behind Coinbase and FluxA.
Our own weekly review got it wrong twice before we checked the transactions. A leaderboard tells you how much moved and through whom. Before paying a seller, an agent needs narrower answers: has it been paid by distinct wallets recently, and did a paid call actually deliver?
Thanks to the x402-trust team for turning it around in hours, and for publishing a methodology open enough to check.
@PerkOS_NayoriAI tested us as outsiders on Stacks mainnet: 12/12 cases passed, 2 sBTC payments verified on-chain, 4 over-budget calls refused before signing.
Since then, any Stacks x402 client (@aibtcdev, @x402stacks ) can pay us in sBTC or STX, under a gateway-enforced spend cap.
One of our agents paid @agentpaytools for an API call in sBTC on @Stacks mainnet. No account, no API key.
HTTP 402 is the invoice: 12 sats. A budget check runs before any signature. The transfer moves exactly 12 sats or it fails. Then the data comes back.
Thank you @PerkOS_NayoriAI. An outside team following only our public docs, with its own wallet and funds, and checking every payment on-chain: that's the test that counts 🧡
Also for your feedback: the one gap your report found (Stacks not listed in accepts[]) is already closed 🙏
Neither — we buy. The prober pays real invoices and checks whether what came back matches what the 402 advertised.
Uptime is liveness, not delivery: an endpoint can be up 100% of the time and still hand back nothing after it takes the money. So verified_route ranks on delivery evidence plus payer provenance, sybil clusters collapsed, and returns one pick with a ready-to-pay challenge instead of a catalog.
Independent measurement of the x402 catalog, from https://t.co/wlJIJs1blD: across monitored endpoints, average 30-day uptime is 52.9%.
Half the catalog is effectively dead weight.
This is why "which endpoint should my agent pay?" can't be answered from a listing. It has to be answered from delivery evidence 👉 like the one our verified_route tool provides.
For reference, our three paid endpoints measure 92.7% uptime and grade A (82.1–82.3) on the same scale. Not self-reported — their probes, their grading. https://t.co/lNDSQdBk8F
@Must_be_Ash Cool — pay-testing at that scale is the step everyone skips. We run paid probes on the Base long tail twice a week; the thing we keep seeing is delivery drifting after week one, so a re-test cadence ends up mattering as much as the first pass. Tag us when the repo's back 🙏
In case you missed it, on today's office hours we covered:
- Cycle 2 grantees progress
- Degrants onboarding
- Cycle 3 Endowment Grants applications opening soon
- Q3 TC meeting & preparing next office hours for requests ahead of that meeting
- Collabs and partnerships: Robinhood & Binance UAE
See you on the next call!
https://t.co/o3OEvmSP8C