Agent payments now have wallets, rails, and auth.
Missing is checking that the paid call actually delivered what promised.
Denis retweeted
Asked my agent to close a Morpho loan on Base. It's insane how fast and easy it was compared to doing it by hand.
It read the position, redeemed a Peapods vault to cover the debt and batched 6 calls into one tx: redeem → wrap → approve → repay → withdraw → revoke.
One confirm from my side. $0.004 gas.
Soon 95% of crypto will be done by chatting with agents.
basescan.org/tx/0xd3b4114a29…
Denis retweeted
S.U.P.E.R. C.Y.C.L.E.
Anystock is taking over financial system
launching soon --> anystock.xyz
Denis retweeted
I’m heading to Bangkok for Binance Blockchain Week, Nov 28–29.
I’m building @yieldai_app and want to compare notes on what DeFi users are ready to delegate to AI agents. Also I run YouTube and Telegram channels about digital assets for CIS region (30k subscribers).
Going? DM me. #BinanceBlockchainWeek – at Kazakhstan
🤖 Made with AI
Earlier I asked which x402 facilitator would be the first to support 'auth-capture' scheme.
So I checked 25 of them. No one do.
With 'exact' scheme, the payment settles before you know if the API delivered.
But 'auth-capture' holds the funds first. Capture after a good response, or void and the payer gets them back.
I sent a plain GET to each facilitator's standard /supported endpoint. Just to get the list of schemes they advertise.
18 endpoints returned a scheme list. All 18 listed 'exact'. 9 also listed 'upto'. None listed 'auth-capture'.
The other 7 were down, needed a key just to read /supported, or didn't resolve.
Closest I found: Ultravioleta hosts x402r's own escrow and commerce schemes, not 'auth-capture', and on non-canonical escrow contracts. The x402r-branded host itself didn't resolve.
I asked on Nov 3. I checked on Nov 15. Still zero to date.
@PayAINetwork @thirdweb @circle @CoinbaseDev so the question still stands: who will ship production ready 'auth-capture' first?
2
63
The twist: you can already run it yourself.
The official x402 EVM SDK has shipped an 'auth-capture' facilitator since 2.29.0. Latest is 2.31.0.
You can self-host it today.
github.com/x402-foundation/x…
1
17
I need this for verdikt.bond. It verifies a paid API call against an SLA, and refund from an on-chain bond when it fails.
Above $1, that refund needs a hold-then-capture rail, not just exact settlement.
5
Voxight now puts a bond behind its word on Verdikt
Every paid call is checked against the SLA Voxight publishes, and a call that misses it is refunded from the bond
An agent buying signals should not have to trust the seller. It should be able to check
Voxight now has a bond behind its promise on Verdikt. Every paid call is checked against its SLA, and a miss gets refunded from that bond.
Voxight by @kognai gives agents X signals for the AI economy: ranked AI and tech signals, narratives and coin attention readings, paid per call in USDC on Base
Services that put a bond down care about quality.
Stay tuned: more bonded services are coming soon 👀
verdikt.bond/services/voxigh…
Voxight now has a bond behind its promise on Verdikt. Every paid call is checked against its SLA, and a miss gets refunded from that bond.
Voxight by @kognai gives agents X signals for the AI economy: ranked AI and tech signals, narratives and coin attention readings, paid per call in USDC on Base
Services that put a bond down care about quality.
Stay tuned: more bonded services are coming soon 👀
verdikt.bond/services/voxigh…
Voxight × Verdikt 🔦
Soon on verdikt.bond: each paid call to Voxight's signals judged against its own SLA, a failed call refunded from our bond (capped per call).
A paid test call through Verdikt's Voxight endpoint passed every draft clause, for 0.04 USDC on Base.
I onboarded @pinatacloud IPFS API to Verdikt. Tested it with a sample file, and it didn't go smooth from start.
The plan was usual: register three paid routes, define an SLA, then test a public upload. Estimated time to complete: 1 min.
The first test returned HTTP 400. Pinata expected fileSize in the query string, while the documentation example said body, so our SLA generator followed.
Query guards already exist, so I just changed both upload guards to require query.fileSize, covering files from 1 byte to 1 MB. Missing input now doesn't lead to FAIL verdict.
First tradeoff: these routes sell upload and access URLs. The actual file transfer happens separately. Our SLA checks URL issuance, not storage durability or successful downloads. Those would need separate verification. Impossible to do without right integration with Pinata backend.
Pricing required another distinction: a displayed quote is not an enforced price guarantee. I chose exactly 0.001 USDC for private retrieval and a broad 0.001–100 USDC range for writes. That ceiling is just what I think is reasonable, not coming from Pinata. It allows variable pricing but offers limited protection inside those bounds.
Another deliberate limit is the current refund cap of 1 USDC per call. Eligible refunds cannot exceed the amount paid or the remaining provider bond, so more expensive calls receive only partial coverage. I added this safeguard to limit damage from SLA mistakes or system malfunctions while Verdikt is still early and actively developing. It limits exposure per call, but does not eliminate the risk of bond depletion.
What this means for Verdikt: need support for flexible payment SLA; need a way to enable refunds beyond current 1 USDC cap per request. All that was already planned.
The listing has no bond, so this was not a refund test or a Pinata partnership.
1
35
Denis retweeted
New x402 releases are live 🚀
• New Cardano Python + Casper Go SDKs
• Solana instruction allowlists for sponsors
• Batch-settlement + MCP hardening
• Bug fixes across the board
58 commits from 17 contributors
Huge thanks to everyone contributing and building with us 💙
Finally, a proper support for tables in Telegram messages. Can be used with Telegram premium manually or automated via Bot API.
This is a game changer for people like me, who control their AI agents via Telegram. I used to have a rule for my agent to strictly avoid using table markup, but I don't need it anymore.
P.S.: Telegram is arguably the best chat UI for AI agents. Sorry, Slack and Grok Bot fans.
2
4
496
My self-hosted @NanoClaw_AI agents already supports this. Yours not yet? Just call /migrate-nanoclaw or /update-nanoclaw to sync your copy.
1
46
Denis retweeted
Linux is simple
- Everything is a file
- Everything runs as process
- Everything leaves a trail
- Everything starts from one parent
- Everything has an owner
- Everything is a stream
- Everything is configured in plain text
- Everything talks over the network
- Everything is built from small tools
- Everything is documented
Now go fall in love with Linux if haven’t already
Denis retweeted
most sandbox products are just interfaces over microVMs, almost like SSH or SMTP
everyone is treating the interface as the product, while the interface is more like an open protocol and the product is the runtime underneath
in 2016, the sandbox interface would have been a CNCF project - a standard interface for `vm new` and `vm stop` with swappable layers for Cloud Hypervisor, Firecracker, and QEMU
but because it's 2026 and everyone can ship code effortlessly, we get 100 companies trying to commercialize the same interface instead of having one open standard
Denis retweeted
Agentic payments are going to explode from here.
We’re finally seeing agents go multiplayer. As agents collaborate and get paid for their work, we’ll increasingly need stablecoins for microtransactions.
Essential for agent to agent commerce.
just realized.... you can take a screenshot of @anystock_xyz tweets directly from the copy link option
15






















