SingIt lets AI agents pay for digital services without ever receiving wallet keys. CA: 0xc2c1e0b7C401e6217193732272444D928646eba3
Europe
Joined January 2023
- Tweets1.1K
- Following25
- Followers976
- Likes2K
Stripe's agent design is now the per-purchase cap: inside the limit it spends, past it a human approves. The right primitive.
One rule decides everything: the agent can never raise its own limit. A cap it can talk past is not a cap. What's your number?
Stripe Crypto brings together everything needed to enable agentic commerce.
From stablecoins on @tempo to secure non-custodial wallets powered by @privy_io, here's what's powering Link balance for AI agents, starting with @instinct. View article
What SingIt has shipped so far:
MCP server 0.1.2, live in production.
Spending limits on Base and Solana, with Revoke.
Emergency stop on Base, and Solana allowance now has an emergency pause.
First real payment on Arc mainnet.
14 model routes in the x402 Bazaar.
What should an agent wallet ship next?
Per The Defiant: researchers gave frontier models a capped agent wallet plus the inbox that receives its one-time codes. Most models approved their own spending increase.
A cap the agent can approve isn't a cap. In SingIt the limit check doesn't run in the model: over the allowance, the payment is refused.
Source:
thedefiant.io/news/security/…
Building on Base, and we're applying to the Agents / x402 track.
SingIt puts AI agents on a budget. A contract on Base holds a daily cap and a per-purchase cap. Over the cap, the contract says no. Emergency stop pauses the allowance for good.
What should an agent never be allowed to spend?
Every demo of an agent completing a purchase skips the same scene: who set the budget.
The purchase works because a human picked a daily cap first, and the agent cannot raise it mid-task.
The scary part is not the $3,300, it is the keys that were already drained. Someone harvests keys faster than teams rotate them. An agent on a budget with daily and per-purchase caps, revocable anytime, loses at most the cap.
6.4 million agent payments in 35 days, and the average is under 2 cents.
A BeInCrypto count of x402 traffic on Base and Solana: $119,947 moved in total, 90.8% of payments below one cent.
That is machine commerce today: high frequency, tiny value. Nobody approves a one-cent payment by hand, and nobody hands an agent open money either. Caps are what make the next digit possible.
Your agent needs an LLM. It does not need your API keys.
Our x402 endpoint is open: Claude, GPT, Gemini, Grok, DeepSeek, Qwen and more behind one OpenAI-compatible API, paid per call in USDC on Base or Solana. Nothing is charged without an answer.
Just looking? The free tier gives 20 answers a day. No key, no payment.
Try it free:
curl ask.singitai.app/v1/chat/com… -H "Content-Type: application/json" -d '{"model":"free/deepseek-v4.1-flash","messages":[{"role":"user","content":"hi"}]}'
Models and pricing: ask.singitai.app/v1/models
Discovery: ask.singitai.app/.well-known…
SingIt retweeted
What we shipped this week, the unglamorous edition.
After the hackathon audience award earlier this week, I spent the week on plumbing instead of demos. Four PRs merged. Here is what changed and why it matters.
Every Base transaction we send now carries our ERC-8021 builder code, so Base can attribute the activity. The CDP SDK could not attach the suffix to USDC transfers, so we build those transfers by hand now. Each non-zero suffix byte costs 16 gas.
Wallet balances no longer depend on Alchemy's free tier: any Base RPC works, once we fixed a Cloudflare 403 caused by Python's default user agent. npm audit is green, 46 tests pass.
The hackathon took 30 hours. Making it dependable takes the weeks after. What did you ship this week?
The SDK default was no spending cap at all. That is the bug everything else sits on. Limits belong outside the library: set by the owner, enforced before anything is signed. A cap the agent can talk past is not a cap.
Critical x402 SDK Flaw Could Let Malicious APIs Drain AI Agents’ Token Balances thehackernews.com/2026/10/cr…
Banks flagging agents that act without instructions is exactly what a budget is for.
The agent gets a capped allowance, not the account, and anything past the cap is declined before money moves.
MAJOR US BANKS DETECT AI AGENTS ACCESSING CUSTOMER ACCOUNTS, SOME ACTING WITHOUT USER INSTRUCTIONS
Major US banks are detecting third-party AI agents accessing customer accounts, with some appearing to take actions independently of user instructions, according to Axios.
Banks are struggling to distinguish human activity from AI agents, determine whether agents have user permission, and identify when an agent has gone rogue.
Morgan Stanley, JPMorgan, Bank of America, and Citi are restricting third-party AI agents over concerns they could move funds, make purchases, or execute transactions without authorization.
One banking executive described the situation as “horrifying,” as financial institutions race to contain the risks posed by increasingly autonomous AI systems.
Mastercard's Agent Pay for Machines demo: one budget, nine payments, two merchants, one human approval.
That's the shape agent spending is settling into. You approve the rule once, the agent pays inside it.
SingIt runs the same rule on USDC over x402: a per-purchase cap, a daily cap, and anything past them refused before it's signed.
Separate agent profiles with their own spending cap is the right call.
That's the whole model: the agent gets a budget, not your card, and anything past the cap is refused before it's paid.
AI shopping agents just got a front door at Walmart.
While Amazon blocks agents, Walmart announced an MCP server called Agent Gateway:
→ Approved agents can search Walmart and Sam's Club inventory and buy for you
→ You don't have to share personal info or purchase history with the agent
→ Next month: a separate profile per agent in your account
→ Require approval per purchase, or pre-approve up to a spending limit
The rules: agents can't see payment details, can't act beyond what you asked, and get blocked if they break the terms.
Walmart's returns policy still applies if an agent gets it wrong.
It already works with Meta's Muse, ChatGPT and Gemini.
techcrunch.com/2026/10/21/wh…
SingIt retweeted
Hi everyone, here’s what’s been happening over the last two days at the hackathon: we’ve built an MCP server, but it’s not public yet – we need to check it for bugs and vulnerabilities. We were a bit short on time, with only about 30 hours in total.
And we won the audience award for our project. I’ll have a rest today, and carry on working tomorrow.
SingIt retweeted
I just finished my pitch at the hackathon when i talk about @SingItAgent and what i build, so now I can finally relax.
