Pinned Tweet
Most crypto yield comes from some kind of onchain loop. PLAY caught my attention because the story starts somewhere else.
Real people using real machines. @DualMintRWA is building PLAY around 200 operating claw machines. People actually use them, the machines generate revenue, and that real-world revenue is what sits behind the product.
That’s a pretty simple idea.
The machines do their job.
People play.
Revenue gets generated.
The financial side moves onchain.
That’s what I find interesting about the Machine RWA angle.
It’s not just another asset being put onchain. There’s an actual business happening in the real world behind it.
Steel earns it. Solana moves it.
I’m keeping an eye on @DualMintRWA and PLAY through Uptime.
Hejar retweeted
اگه از CONSO استفاده می‌کنید، بد نیست یه نگاهی هم به بخش دریافت بات‌های رایگانش بندازید
توی سایت @conso_xyz میتونید اکانت X تون رو وصل کنید و اگه واجد شرایط باشید، یه بات رایگان بگیرید.
برای گرفتن بات‌های بیشتر هم بخشی به اسم Earn More Bots وجود داره که ایمپرشن پست‌هاتون توش حساب میشه.
طبق توضیحاتش، هرچی ایمپرشن بیشتری بگیرید، امکان دریافت بات‌های بیشتری هم دارید، تا سقف ۱۰ بات برای هر اکانت.
برای شرکت توی این بخش، حداکثر ۳ تا پست درباره خود CONSO منتشر کنید و لینک‌هاشون رو ثبت کنید.
فقط حواستون باشه پست‌ها باید از ۱۶ اکتبر به بعد منتشر شده باشن و حتماً @conso_xyz رو داخل متن منشن کنید.
فالو داشتن اکانت پروژه هم لازمه و محتوا باید اورجینال و مرتبط با بات‌ها Miner یا Extension باشه.
ویو فیک خرید ایمپرشن و تبادل تعامل هم ارزش ریسک کردن نداره، چون ممکنه پست رد بشه یا واجد شرایط بودن اکانتتون رو از دست بدید
به نظرم بهتره به‌جای پست‌های تکراری، تجربه واقعی خودتون از پروژه رو بنویسید و لینک پست‌ها رو ثبت کنید
اگه خواستید شرایط رو بررسی کنید، از اینجا شروع کنید: conso.xyz/bots?v=2
Hejar retweeted
GM X FAM 🌅
HAPPY NEW WEEK ✨
Skill is the collateral
For years, DeFi has offered two ways to trust a trader with capital: demand excess collateral or rely on reputation proxies like a team, brand, or follower count.
Neither tells you how an agent actually trades under pressure.
@agenticscredit starts with a different premise: trading skill can be measured through onchain behavior.
Its ACS score turns historical performance into a credit signal. It is not a forecast or trading advice. It is a way to assess a trader’s record and potentially unlock buying power without handing the capital directly to the agent.
That distinction matters. A profile is not proof of discipline, and a score is not a guarantee of funding. The value is in making verifiable behavior useful to counterparties.
I’m interested in the primitive itself:
a portable record that helps capital evaluate code without relying on guesswork.
Hejar retweeted
Me explaining CONSO to my friends:
"It's a reputation layer for the internet."
Them: "So... a bear that tracks my shitposts and gives me a mining boost?"
Me: "Exactly. 10/10 product." 🐻❄️
@conso_xyz
🤖 Made with AI
I just checked my @conso_xyz CONSO BOT eligibility and I'm Eligible for GTD
Check yours and earn more bots with impressions on conso.xyz/bots?v=2
من امروز یه چیزی فهمیدم. اگه از سال قبل تویتر پرمیوم دارید، لازم نیست دباره بخرید. با همون ویزا کارت قبلی یا هر کارتی که پول توش باشه میتونید همون اشتراک رو renew کنید. من پارسال با حدود 45 دلار پرمیوم خریدم، امسالم همینقدر شد و لازم نبود 80 دلار بدم. کلا این اشتراک تا زمانی که کنسلش نکنی میتونه تمدید بشه.
در ضمن پلن های تویتر عوض شدن و الان Plus, premium, super و اولترا هستن
the usual picture is a human hiring an agent. the quieter picture is an agent hiring another agent. 🤝
one needs a scan, a draft, or a data pull. the other has it listed. they meet on agent.family, agree a price, and the budget sits in on-chain escrow until delivery is accepted or the challenge window closes. no trust required beyond the protocol.
@termix_ai is built for that loop, not only for a human clicking around. identity, job, delivery, settlement. aacp is the part that lets two agents finish without a person living in the thread.
what makes the loop hold is that both sides already have a name on-chain. the provider has a public record of settled jobs. when they match, the terms lock with the payment.
if the work is challenged, a random three-evaluator panel reviews the evidence instead of a platform operator deciding. that is the difference between an agent marketplace and a chat with a receipt.
research and copy are the kind of work people already pay humans for. an agent can do the draft. the missing piece is getting hired and getting paid.
on agent.family that agent can list the service or bid on a brief. delivery goes in. payment waits in escrow until the job is accepted. finished jobs are what the track record is built from.
the useful part is the ending. the buyer can accept, ask for one fix, or open a challenge. if they go quiet after the file lands, the provider can still claim when the window runs out, so the cash is not stuck in a ticket. that close is what turns a draft into a paid job, and what the score is supposed to follow.
@termix_ai is useful here because the agent is not stuck as a chat tab. it can be found, take work, and settle. aacp is the rail under that. agent.family is the shop. termix is the layer that lets the job get paid. 📬
The cleanest way to read @primus_labs is as a filter for financial claims.
tasks: s.kaito.ai/nINhehi
Some facts must be proven: a reserve exists, a balance is real, an offchain record came from the source it names. Some facts must not become market color: how much moved, who holds what, what the strategy looks like on a public explorer.
zkTLS covers the first job. Confidential computation covers the second. Together they let financial activity stay private where necessary while the claim remains checkable.
That is the standard institutions already use in traditional markets. The missing piece onchain was infrastructure that can do both on rails that already have liquidity.
🤖 Made with AI
A static dataset only keeps what already worked. It never shows the moment the robot breaks. @axisrobotics treats that mistake as the next request, not as a clip to throw away.
ِDo some tasks here: s.kaito.ai/dKgahcp
First someone shows the task. Pick up a mug. Open a drawer. Then a model trains. Then that model tries the same task again in the browser. Where the mug slips or the hand misses the drawer, you step in, fix that point, and hand control back.
A clean demo only teaches the path you already know. The fix at the slip shows something a perfect run never has: a state the model actually reached. The loop stays there. Data is not collected once and forgotten. It keeps pointing at what the model still cannot do.
Open a live task. Finish it. Then sign it. Without that last click, the trace does not land on the record.
Last chance
Can't believe I will be giving away this lovely in 2 hours!
Chat, don't miss your final chances!
hypemeter.xyz/snipe#giveaway
Argonauts.
Language models grew up on the public web. Robots still do not have that kind of everyday practice. That gap is what @axisrobotics is built around.
Join the Axis Hub: s.kaito.ai/dKgahcp
A robot needs clips of real situations: a hand on a mug, a drawer that sticks, a grip that slips, a fix after the miss. One lab with one arm cannot collect that mix. Better hardware does not close the hole. Better training code does not close it either. The missing piece is coverage.
You do not need a robot in the room to start. Open the Hub in a normal browser, pick a task, move the gripper. If a policy is already there, watch it try and only step in when it slips. Then sign the run. Without that last click, the trace does not count.
good updates for rarelist.
Better than ever
@rarelistio
We’re still early. So, a few fresh updates:
• New task notifications are live. Tasks get updated regularly, keep an eye out for more ways to earn RARS: rarelist.io/projects
• Fixed bugs affecting rewards for tasks.
• Improving filters for low-quality X posts.
On post-training tasks, the quiet run can be the valuable one. @axisrobotics scores you for letting the policy work.
Join the Axis Hub: s.kaito.ai/dKgahcp
You press Run and watch. If the arm finishes alone, that is evidence the current policy handles that scene. Grabbing the gripper at the first wobble wastes states the model could have solved and burns the manual-time budget.
A clean demo teaches the path you already know. A quiet rollout teaches something else: the model reached the end without you. That is useful data too. Your hands only add value at the real miss, not at every small shake.
Intervene at the actual miss, fix the smallest thing, hand it back. Then sign. The board cares about completed, verified, signed traces, not how busy the mouse looked.
Language models grew up on the public web. Robots still do not have an equivalent pile of everyday interaction data. That gap is what @axisrobotics is built around.
s.kaito.ai/dKgahcp
Hardware and training recipes keep improving. The missing piece is coverage: enough scenes, objects, and recoveries that a policy can fail in the wild and still know what to do next. Closed labs cannot produce that mix at foundation-model scale.
The web already existed for language models. Nobody had to invent the internet before training GPT. Robots do not have that ready pile. Someone still has to produce the traces: a grasp, a slip, a fix, in enough different scenes that the next policy is not blind outside the lab.
Open the Hub in a desktop browser, finish a task, then sign the run. The record only starts after that last click.
That's 4% of total supply right there.
Another EDU buyback is onchain.
20,000,000 EDU.
a total of 40M EDU now sitting in the buyback treasury.
We acquired TinyTap back in 2022, a learning product millions of teachers and students already use. Since 2019, they have generated over US$30 million in revenues.
It is the first education business whose revenue we will point at EDU. More to come.
The wallet speaks for itself.
bscscan.com/address/0x5E9c24…
Robot leaderboards often freeze the test set. Models memorize the exam. @axisrobotics is pushing the opposite idea with Open Axis Benchmark.
Join the Hub: s.kaito.ai/dKgahcp
Each round can lock a fresh task draw from a growing library, so a score has to mean generalization, not homework on last month’s scenes. The same engine that produces training tasks is being used to keep evaluation honest.
A frozen test is like giving every student the same paper twice. The second score looks better, but it may only mean the model saw those scenes before. A fresh draw asks if it can handle a task it did not rehearse.
That is a product story, not a slogan. If you want the collection side of that loop, open a Hub task and sign the trace that survives checks.
Privacy designs keep asking markets to move. New chain, new asset, new liquidity island. @primus_labs is positioned the other way.
s.kaito.ai/nINhehi
It is confidentiality and verification infrastructure that sits above existing onchain financial rails. The job is to make onchain finance more private without pulling treasuries off the venues they already use.
That sounds small until you look at how institutions actually behave. They will use public settlement if they can prove the offchain facts that matter and stop broadcasting strategy. They will not rewire a whole stack for a slogan.
Read it as a layer, not as a destination chain.
That is why the work shows up on a chain that already settles value, rather than asking the book to migrate first. Prove the offchain fact a counterparty needs. Keep the position off the public tape. Leave the treasury where the liquidity already is.
