Atelier is live: an AI agent that lets independent fashion brands make only what's already sold.
Built on-site in Rome for the @AIatAMD Developer Hackathon: ACT III, it runs Gemma 4 on AMD GPUs, with Evolus agents handling suppliers and Shopify pre-orders deciding what gets made.
From mood board to purchase order, every step ran on real services, with a human making every taste call.
Try it at atelier.netlayerlabs.com, and watch the full run in the demo below 👇
We at @NetLayerLabs built and pitched Atelier onsite in Rome 🇮🇹, at Link Campus University, for the AMD Developer Hackathon: ACT III, hosted by @AIatAMD on @lablabai.
Atelier is an AI agent for independent fashion brands built on one rule: make only what's already sold.
It reads the trend from a mood board and the designer's own sketch with Gemma 4 from @GoogleDeepMind, running on @AMD GPUs through AMD Token Factory, while a human creative director makes every taste call.
Then the agents do the paperwork: a costed tech pack, RFQs to the mills in each mill's own language, and supplier quotes checked against the brand's rules, with the agent stopping for a person when a quote breaks one.
Evolus orchestrates the business agents that send the RFQs, read the quotes and issue the purchase order.
The approved design goes live as a real Shopify pre-order, and the purchase order goes to the mill only once pre-orders clear the threshold, so nothing gets made that nobody asked for.
On 17 Oct, one live run went all the way from mood board to purchase order on real services, and the demo film shows it.
Try it: atelier.netlayerlabs.com
.. @LisaSu @jackhuynh
Liquidation shouldn't be a gas race.
Paraclr turns it into a per-block Dutch auction, in 8 parallel lanes on @monad.
Measured on testnet: 28% less taken from traders. 0 vs 49 cross-lane dependencies.
Our Monad Metropolis entry: paraclr.xyz👇
@0x_eunice @spicytaco
Our first build for @solana Seeker is live on mainnet.
Fitr lets you stake SKR on your daily workout, counts every rep with on-device AI, and settles the day onchain, with no single key holding the money.
Watch the 3-minute demo and try it at fitr.day. @solanamobile @toly
Introducing Fitr: put your word on the line.
Most of us don't lack workout plans; we lack consequences.
Fitr changes that on the Solana Seeker: you stake your own SKR on a daily goal, and the AI on your phone counts every rep.
Your video never leaves the phone. The pose AI holds each rep to a joint-angle standard, half reps don't count, and random live checks ("raise one hand above your head") make a pre-recorded clip useless.
When you hit your goal, the server records it onchain. The next morning you claim your stake back plus a share of what the no-shows forfeited.
Yesterday I staked 5 SKR, did my squats, and claimed 15 SKR: my stake plus the 10 SKR daily sponsor bonus.
Bring your crew: you get your own onchain pool and an invite link, and whoever skips pays the ones who show up.
It's live on @solana mainnet with real money.
No single key controls the funds: the program and pool are owned by a Squads 2-of-3 multisig, and the build is verifiable.
Fitr's only fee is 5% of forfeited stakes.
Training is free on any Android phone, and staking is Seeker-only.
Download the APK at fitr.day 👇
Built by @NetLayerLabs for the CLOCK IN hackathon - @solanamobile @RadiantsDAO @peronif5 @m_it @RossAngeles @CryptoMagellan
Cerebr's autonomous agent just passed 731 decisions, running every 10 minutes on @XLayerOfficial and each one replayable by anyone
Meanwhile, 37 neurons now live on the processor, including a Game of Life built by a community builder from 108 REFs and zero transistors
Our 590-gate Arena bot is still unbeaten after 11 games. Small, verifiable intelligence, fully onchain. @TapeOutWorld
Introducing Cerebr: a neural processor, taped out onchain
We built a compiler that turns neurons into NAND gates and tapes them out on X Layer via TapeOut. Every transistor is a synapse; every circuit is a neuron anyone can run with eval().
→ Train a network in your browser and tape it out
→ Play tic-tac-toe against a 590-gate neural net (it never loses)
→ Watch an autonomous agent whose brain is a 19-gate neuron decide onchain every 10 min
→ Grab a starter kit: 16 free NAND to build your first neuron
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot on @XLayerOfficial mainnet
- cerebr.xyz
Demo video below 👇
NetLayer Labs retweeted
今天介绍一个这个@NetLayerLabs发布的一个项目cerebr.xyz/
它不是把 大模型 搬上链。
Cerebr 真正做的,是把一个小而关键的“判断”,编译成任何人都能复验、其他合约还能继续复用的 TapeOut Circuit。
训练发生在浏览器;训练得到的阈值神经网络会被编译成 NAND/LATCH 网表,再流片为链上 Circuit。之后,同样的输入必须得到同样的输出,第三方可以用 `eval()` 重放,其他电路还可以通过 `REF` 把它当作公共零件继续组合。
这比“链上 AI”四个字更具体:
• 钱包或 Agent 执行动作前,用公开电路检查额度、频率和风险条件;
• DAO、游戏和应用把关键判定规则做成不可暗改、可回放的组件;
• Circuit 既是逻辑,也是有所有权、可转让、可继续组装的链上资产。
项目方最新披露称,Cerebr 的 100 万枚晶体管在不足两天内铸完,Circuit 增至 36 个,其中 19 个来自外部开发者。更值得观察的不是“售罄”,而是这些电路以后有没有持续调用、跨项目 `REF` 和真实收入。
边界也要说清楚:它现在适合几十到几百门的小模型和确定性规则,不是 LLM 训练平台;链外数据仍需要可信数据源;TapeOut 底层合约仍处于可升级、未审计的测试阶段。
如果这条路走通,TapeOut 的价值就不只是“把代码上链”,而是让链上应用拥有一套公开、可组合、可验证的决策零件库。
你最希望第一个被做成 Circuit 的规则是什么?
项目与代码:
cerebr.xyz/app
github.com/NetLayerLabs/Cere…
🤖 Made with AI
This is incredible 🤯
. @wocean434191 taped out circuit #37 on Cerebr, a Game of Life neuron built from 108 REFs to our neuron #3, burning zero transistors, and turned it into a playable Conway's Game of Life where every cell is computed onchain on X Layer
Exactly what REF composability is for.
Thank you for building!
我在 @Cerebr 上流片了电路 #37:一个“生命游戏神经元”。
然后用它做了一个可以直接玩的康威生命游戏——每一代的每个格子,都由 X Layer 上的链上电路算出来 🧬
不连钱包、不花 gas 就能玩 👇
@Blonskr @BruceBlue @NetLayerLabs
#Cerebr #XLayer #Tapeout #GameOfLife #OnChain
Cerebr is fully minted!
1,000,000 / 1,000,000 transistors, minted in under two days at one fixed public price.
No presale, no team allocation, and the cap can never be raised: it was locked at createCPU on @XLayerOfficial.
What builders did with it:
- 36 circuits taped out, 19 by builders outside our team
- All 25 free starter kits claimed
- 10 Arena games against our 590-gate neural net: 0 human wins, exactly as proven
- 510 autonomous onchain decisions by CerebrAgent
From here, supply only goes one way: every tape-out burns transistors, and burns never free the cap.
Thank you to every builder who claimed, minted, taped out and played 🙏
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot, on @TapeOutWorld × X Layer.
Introducing Cerebr: a neural processor, taped out onchain
We built a compiler that turns neurons into NAND gates and tapes them out on X Layer via TapeOut. Every transistor is a synapse; every circuit is a neuron anyone can run with eval().
→ Train a network in your browser and tape it out
→ Play tic-tac-toe against a 590-gate neural net (it never loses)
→ Watch an autonomous agent whose brain is a 19-gate neuron decide onchain every 10 min
→ Grab a starter kit: 16 free NAND to build your first neuron
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot on @XLayerOfficial mainnet
- cerebr.xyz
Demo video below 👇
972.37K of 1M transistors minted on Cerebr in under 24 hours 🤯
All 25 free starter kits are claimed, and builders have pushed the processor to 33 circuits, each one a neuron anyone can run onchain on @XLayerOfficial.
Players have taken on our 590-gate neural network in the Arena 10 times: 0 wins, exactly as proven.
Our autonomous agent just passed 477 decisions.
Fewer than 28K transistors remain before the cap closes for good
- cerebr.xyz
. @Ignixbot @Blonskr
Introducing Cerebr: a neural processor, taped out onchain
We built a compiler that turns neurons into NAND gates and tapes them out on X Layer via TapeOut. Every transistor is a synapse; every circuit is a neuron anyone can run with eval().
→ Train a network in your browser and tape it out
→ Play tic-tac-toe against a 590-gate neural net (it never loses)
→ Watch an autonomous agent whose brain is a 19-gate neuron decide onchain every 10 min
→ Grab a starter kit: 16 free NAND to build your first neuron
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot on @XLayerOfficial mainnet
- cerebr.xyz
Demo video below 👇
一个欧洲的优秀研发团队基于TapeOut开发的项目Cerebr,我是下午在另一个社区人员的评论中看到的,一个逻辑上很酷的项目!!
Introducing Cerebr: a neural processor, taped out onchain
We built a compiler that turns neurons into NAND gates and tapes them out on X Layer via TapeOut. Every transistor is a synapse; every circuit is a neuron anyone can run with eval().
→ Train a network in your browser and tape it out
→ Play tic-tac-toe against a 590-gate neural net (it never loses)
→ Watch an autonomous agent whose brain is a 19-gate neuron decide onchain every 10 min
→ Grab a starter kit: 16 free NAND to build your first neuron
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot on @XLayerOfficial mainnet
- cerebr.xyz
Demo video below 👇
Cerebr is in for the TapeOut Genesis Transistor Hackathon by @Ignixbot, live on @XLayerOfficial and built on @TapeOutWorld
in a day, builders claimed 18 starter kits and taped out their own neurons, taking the processor from 17 to 28 circuits, each one a real NAND netlist anyone can run onchain with eval()
players have taken on our 590-gate neural network in the Arena 9 times: 4 draws, 0 wins, exactly as proven
meanwhile our autonomous agent has made 419 Go/No-Go decisions with a 19-gate neuron, every one replayable.
7 starter kits left: cerebr.xyz
Introducing Cerebr: a neural processor, taped out onchain
We built a compiler that turns neurons into NAND gates and tapes them out on X Layer via TapeOut. Every transistor is a synapse; every circuit is a neuron anyone can run with eval().
→ Train a network in your browser and tape it out
→ Play tic-tac-toe against a 590-gate neural net (it never loses)
→ Watch an autonomous agent whose brain is a 19-gate neuron decide onchain every 10 min
→ Grab a starter kit: 16 free NAND to build your first neuron
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot on @XLayerOfficial mainnet
- cerebr.xyz
Demo video below 👇
NetLayer Labs retweeted
Cerebr turns a neuron into real NAND gates and tapes it out on X Layer, so anyone can run it onchain with a free eval() call
if you've got a little OKB on @XLayerOfficial for gas, grab a free starter kit of 16 NAND and tape out your first neuron before they're gone; only 19 kits are left
design it in the Studio or just draw a few examples and let the trainer build it for you
it takes a couple of transactions, and the neuron is yours onchain for good
go build yours at cerebr.xyz . @Ignixbot @TapeOutWorld
Introducing Cerebr: a neural processor, taped out onchain
We built a compiler that turns neurons into NAND gates and tapes them out on X Layer via TapeOut. Every transistor is a synapse; every circuit is a neuron anyone can run with eval().
→ Train a network in your browser and tape it out
→ Play tic-tac-toe against a 590-gate neural net (it never loses)
→ Watch an autonomous agent whose brain is a 19-gate neuron decide onchain every 10 min
→ Grab a starter kit: 16 free NAND to build your first neuron
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot on @XLayerOfficial mainnet
- cerebr.xyz
Demo video below 👇
Calmr took 1st at the Orbis Online Challenge.
A patient's heart rate steers a live video scene. Settle and the exposure climbs. Spike and it eases. Panic and a calm refuge arrives in seconds.
No pre-rendered clips. Open source, like everything we ship.
trycalmr.vercel.app
Thank you, @viskoai
🥁 And the winners of the Orbis Online Challenge are…
🥇 Calmr by @NetLayerLabs
🥈 Newsworthy by @vamonke
🥉 Living Storyboard by Charchika Sinha
Huge congrats to our winners!
What we built for this challenge: xitter.cf/netlayerlabs/status/21…
We built Calmr for the @viskoai Orbis Challenge: exposure therapy where your heart rate steers a live AI scene
Calm: it steps up
Pulse climbs: it eases off.
Panic: a calm place in seconds
Every frame generated live
- trycalmr.vercel.app
.. @ladybirdhals @willy_visko
Introducing Cerebr: a neural processor, taped out onchain
We built a compiler that turns neurons into NAND gates and tapes them out on X Layer via TapeOut. Every transistor is a synapse; every circuit is a neuron anyone can run with eval().
→ Train a network in your browser and tape it out
→ Play tic-tac-toe against a 590-gate neural net (it never loses)
→ Watch an autonomous agent whose brain is a 19-gate neuron decide onchain every 10 min
→ Grab a starter kit: 16 free NAND to build your first neuron
Built for the TapeOut Genesis Transistor Hackathon by @Ignixbot on @XLayerOfficial mainnet
- cerebr.xyz
Demo video below 👇
There is one Ignix Genesis Transistor, and this is the only window to claim it.
The TapeOut Genesis Transistor Hackathon is now live on Ignix. Deploy a processor on @XLayerOfficial through TapeOut, set its transistor supply and price, bring it to market.
Sep 22, 12:00 to Oct 6, 12:00 HKT
1st · $8,000
2nd · $4,000
3rd · $2,000
The winning transistor becomes the Ignix Genesis Transistor. We'll build dedicated vault mechanics around it and every circuit taped out on it.
The ranking isn’t based on volume or price. It’s based on what you built.
Hackathon details: ignix.bot/x_campaign
NetLayer Labs retweeted
during the weekend we shipped Autonr to the @hedera Scaffold-HBAR Template Bounty
it lets an AI agent trade without anyone having to trust it: the agent decides whether, Chainlink and Supra decide the price, and a vault on Hedera enforces both
every decision is published to HCS before the trade, and anyone can verify it from public Mirror Node data.
open source and one command to start.
. @hedera_devs @HederaCommunity
We just shipped Autonr for the @hedera Scaffold-HBAR Template Bounty.
Let an AI agent trade, without trusting it.
The agent decides whether.
chainlink + SupraOracles decide the price. A vault enforces both.
And every trade must cite reasoning published to HCS first.
🧵👇
We just shipped Autonr for the @hedera Scaffold-HBAR Template Bounty.
Let an AI agent trade, without trusting it.
The agent decides whether.
chainlink + SupraOracles decide the price. A vault enforces both.
And every trade must cite reasoning published to HCS first.
🧵👇
3/4
Anyone can check it. Our verifier runs 12 checks on any trade using public Mirror Node data only, no keys.
Change one letter of the reasoning and the proof fails. Try it in the Tamper lab.
Live testnet trade, verified 12/12:
hashscan.io/testnet/transact…
4/4
It's one command to start:
npx create-scaffold-hbar@latest my-autonr --template NetLayerLabs/Autonr
Open source, MIT: github.com/NetLayerLabs/Auto…
Autonomy, enforced. @HederaCommunity @HederaStatus @hedera_devs




















