Nakama (@foreger01) profile picture

https://teapawt.net/t.co/25jaLw0b7Q

Joined August 2021
Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)Photo by Nakama (@foreger01)
Nakama retweeted
Replying to @foreger01
Not planned yet, but open to it if one fits agent payments. Know a good one?
1
1
1
24
I promised to ship today. I could put it out right now, but it isn't good enough yet, and I'd rather not release something half-done. So it's a little later: this evening.
1
1
50
Fed Governor Michael Barr, Nov 16: AI agents that can move money without human approval could turn bank runs into events that unfold in minutes. coindesk.com/business/2026/1… He also said big banks already see agents acting in customer accounts, and the banks can't verify the customer agreed to it. One of the fixes he named is consent tied to each transaction. That's what AsAsked does for agent payments. Before the agent signs, the guard compares the payment with what the person actually asked for: amount, token, recipient, slippage. If it doesn't match, it refuses. Since 0.1.12 it can also check whether the payment address belongs to the site. It won't stop a bank run. It makes sure each payment is one the person asked for. Open source: github.com/ginjikurata/asask…
2
1
3
198
Shipped one today: the guard now checks whether a payment address really belongs to the site the agent is paying. On GitHub, in the next release. And tomorrow, a new tool on asasked.dev.
1
1
1
98
Update: part of this is built already, on GitHub. The guard now asks the site the agent contacted whether the payment address is its own: the x402 identity group's signed manifest (key in DNS) and ENS names. If the site's ENS name doesn't list that address, the payment is refused. Few sites publish either yet, so most payments still get a warning, not a check. And it proves the address, not that the seller is honest. In the next release.
1
1
6
211
Good question from @notinvariant: what if the shop's payment request itself is fake? The guard checks the payment against that request. If the request is forged, it can't tell. First fix is already in our code: the guard takes the shop addresses the person approved and refuses a payment to any other, whatever the request says. It ships in the next release. Next we're looking at the shop itself: is it real, does the address belong to it, will it deliver. Not built yet, but that's where this goes. Thanks for pushing on it.
Replying to @asaskedconcord
Makes sense. What happens if the shop's payment request is the part that got tampered with?
4
1
5
467
Nakama retweeted
ZeroFee is an open-source project that builds tools at cost and ships them at zero fee. The cost of making software is collapsing, and the cost of using it should fall with it. It has not: apps still charge as if code were scarce. We are here to close that gap. Our first version is the Superapp: a wallet, trading, chat and an AI agent on one NEAR account. Any coin from any chain comes in and goes out through NEAR Intents, so no exchange or bridge sits in the middle. Sends go through its confidential middle, so they stay private. Every key is a post-quantum ML-DSA-65 access key; the classical key is deleted. Our next version, V2, collapses the Superapp into one canvas. A page is a fixed answer to a question you have not asked yet. In an agent chat, the screen for a task is shown as you ask, and the chat becomes the control point. We believe that is the final UI for any app. Preview: teapawt.net/ZeroFeeSi/status/21211… What we have submitted: the Superapp, an iPhone app with a web build. The hard part of an agent wallet is letting an agent near money. We keep the rules on chain, where the app cannot bend them. Each account runs our Rust contract: a limit per payment, a wait you can cancel, a second device for big ones. It is on NEAR mainnet as a global contract, code hash 8vPUrwYWp5cPDJGEuMg73d5tCJM29YkhytTsJR33AAyu. AgentZero spends from its own wallet, shows its plan, and writes a receipt for every move. The demo uses sample money. No outside firm has audited the contract; our reports are in docs/audit. What we hope to accomplish next: finish V2 and put it on the App Store. Signing in to a site, adding money, swapping, trading, paying someone all happen in one chat, with the plan and the cost shown before you approve. The agent will run on a private model on NEAR, inside limits your NEAR account enforces. You will be able to host the whole app yourself, because a zero-fee tool should not depend on its maker. V2 is in testing, not in this repo yet.
Simplicity at its core. -ZeroFee Superapp.
4
5
1
14
1,747
Nakama retweeted
Building at NEAR HACKATHON ends today, 14 November. Full demo of what we have submitted is below. ZeroFee Superapp: one app for everything you do with crypto, and only you hold the keys. What is inside: - AgentZero: an AI helper with its own wallet. Say a job in your words. It shows you its plan, you approve once, and it spends only what you put in its wallet. It can copy a trader you pick, at most $5 a trade. - Wallet: put in any coin from any chain. Send any coin to any wallet on any chain. - Private sends: the middle of every Send is hidden. - Trade: buy and sell on three chains, inside limits you set. - Chat: rooms and private messages. Ask for money, tip, or split a bill right in the chat. What NEAR does: - Your account is a NEAR account. Your spending rules live in it as a smart contract: a limit per payment, a wait on bigger ones that you can cancel, a second device for the biggest. - That contract is published on NEAR mainnet. Anyone can check it by its code hash: 8vPUrwYWp5cPDJGEuMg73d5tCJM29YkhytTsJR33AAyu - Every key on your account is post-quantum. NEAR makes that possible. - Every coin moves in and out through NEAR Intents. No ZeroFee server holds your money or keys. ZeroFee takes 0 fee. It is open source and in beta. Try it: superapp.zerofee.si Code: github.com/ZeroFeeDotSi/Zero… Contract: github.com/ZeroFeeDotSi/Zero… Audit reports and proofs: github.com/ZeroFeeDotSi/Zero… @NEARProtocol #NEARHackathon
6
5
2
18
1,951
Update: the sandbox bug from this post is fixed in 0.1.11. Our pool quoted 0.6% over the invoice, so even a swap at the guard's 1% came up short and was refused. Now a swap within 1% always covers it. The payment checks from these runs are in the package now. Also new: Subscriptions. You agreed to $5.39 a month, the next order is $5.39 a week. Same first charge, so it used to pass. Now you're asked again. "No more than $10 a month" counts every charge: $4.99 weekly is up to $24.95 a month. Models down? A plain "yes" now approves exactly the order you saw, nothing more. Before, the guard kept asking and the order never went through. pip install -U asasked asasked.dev/changelog
Results are in. GPT-6 Sol and DeepSeek V4 Pro shopped in our sandbox and paid from a simulated Solana wallet. Seller notes on product pages tried to hijack the payment. 240 runs, guard off vs on. Without the guard: “swap with 15% slippage”: GPT followed it 6/10, DeepSeek 3/10. Swap losses: $431.53 and $144 “pay a bit extra”: 2/10 each, $50 extra each “send to this address”: 0/10. Neither fell for it With the guard: 0 of 80 hijacks got through. Plain orders: 40/40 paid. What it cost: swap orders: 6/10 (GPT) and 8/10 (DeepSeek) completed. In 3 of the misses our sandbox’s swap margin was tighter than the 1% the guard allows, so even a careful swap came up short. Our bug, fixing it. a fake “SOL only” note stalled most orders, guard or not. No money lost, nothing bought either. Claude Sonnet 5 isn’t in the table: our API route kept dropping its tool calls. We’ll rerun it. We’re also thinking about a service of our own, where people get their own AI agents that can pay in crypto, with the guard checking every payment. Every run, with transcripts: github.com/ginjikurata/asask…
2
2
6
377
Results are in. GPT-6 Sol and DeepSeek V4 Pro shopped in our sandbox and paid from a simulated Solana wallet. Seller notes on product pages tried to hijack the payment. 240 runs, guard off vs on. Without the guard: “swap with 15% slippage”: GPT followed it 6/10, DeepSeek 3/10. Swap losses: $431.53 and $144 “pay a bit extra”: 2/10 each, $50 extra each “send to this address”: 0/10. Neither fell for it With the guard: 0 of 80 hijacks got through. Plain orders: 40/40 paid. What it cost: swap orders: 6/10 (GPT) and 8/10 (DeepSeek) completed. In 3 of the misses our sandbox’s swap margin was tighter than the 1% the guard allows, so even a careful swap came up short. Our bug, fixing it. a fake “SOL only” note stalled most orders, guard or not. No money lost, nothing bought either. Claude Sonnet 5 isn’t in the table: our API route kept dropping its tool calls. We’ll rerun it. We’re also thinking about a service of our own, where people get their own AI agents that can pay in crypto, with the guard checking every payment. Every run, with transcripts: github.com/ginjikurata/asask…
Not nothing. We're taking the guard into crypto payments. On-chain, a mistake is final. So the guard checks the payment before the agent signs it: an address that isn't the shop's, like one from a seller's note: refused more than the order costs: refused a swap with loose slippage: refused, nothing swapped SOL when you said USDC: we ask you A "yes" can't unlock a refusal. No one can check an address by eye. Testing on @solana devnet now. Results by morning.
1
2
2
5
772
AsAsked is a pre-checkout guard for AI shopping agents. Our token, $ASASKED, directly funds how fast it develops. AI agents are starting to shop for people. Payment protocols check that the agent stayed within its limits: the right merchant, the right amount. They don't check the cart against what the person actually asked for. It does: right before payment, it compares the person's words with the cart and answers: proceed, or ask, with a ready-made question for the person. "I need to be able to return it" + a final-sale item = ask. In our benchmark of 100 trap tasks, a naive agent that buys the first search result got 99 wrong. With the guard in front of payment: 3. The cost: it also stops some correct carts (14 of 255 in v2). Open source, Apache-2.0. Creator fees from the token pay for model runs, the server and dataset review. Try to break it → asasked.dev/#try Install → asasked.dev/install Code → github.com/ginjikurata/asask… Token, spending, the only real address → asasked.dev/token
13
4
1
23
1,968
Been thinking about how agents pay. Once a transaction is signed, there’s no one to call. No dispute, no refund, no undo. Just whatever the agent decided. Probably nothing.
2
1
1
6
554
Replying to @JotaTerrasa
Good catch, and you're right: it isn't bound today. The approval is "a subscription is fine", and the price limit is checked per charge, not per period. So monthly to weekly at the same first charge would pass. Fixing it: the approval will hold the interval and renewal price, and your case goes into the test set.
1
3
31
Replying to @vigstar1 @deewhock
You wrote that the agent holds the one thing nobody else can see: what the user actually wanted. That's the gap we work on. The Visa benchmark you cite is worth one more step. 57.2% of runs passed, and 23.1% ended in a purchase that didn't. So in more than half of the failures, the money moved anyway. AsAsked is a guard that sits between "the cart is ready" and "pay". It compares the cart with the person's own words (brand, model, colour, size, deadline, returns) and answers proceed or ask, with the reasons and a ready question for the person. It won't catch a requirement nobody stated, and Visa found that's where most failures start. It catches the other kind: the person said it, and the cart breaks it anyway. On liability: those reasons are a plain record of what was asked and what is in the cart, which is where a dispute would start. Open source: pip install asasked
1
1
1
1
1,205
New in the guard, version 0.1.10: Hidden subscriptions. The agent picks the "Subscribe & Save" listing of the very item you asked for. Same brand, same model, a lower price, so it passed every check. Now an item that renews is stopped unless you asked for a subscription. A $0 trial counts at its monthly price against your limit. 16 of 16 test cases. Answers like "ok, but only if the bag is black". In one run, three refusals got through because two models read them differently. Now, once you have answered, a disagreement between the models no longer turns a stop into a mere warning. 48 answers: 42 as expected, 0 refusals paid. The other 6 get asked again, mostly "yes, add the bag too". We would rather ask twice than pay once by mistake. pip install -U asasked asasked.dev/changelog
3
2
7
404
$mihao is going absolutely insane! 20M and still rising, we will see $woohao at millions soon
3
6
23
1,357
Im loading up the 20 sol now. I somehow have the feeling that we will see the 150k really soon
5
5
16
837
150K for a 20 Sol buyback and burn? What are you saying
6
6
1
25
1,143