Executive Summary
The past 24 hours have seen significant momentum in the local AI space, with multiple posts demonstrating that frontier-level AI capabilities—including autonomous agentic workflows and computer control—are now achievable on consumer hardware like a single RTX 3090. Hermes Agent from Nous Research has emerged as the dominant framework, with users reporting successful autonomous coding and testing workflows, while OpenClaw faces increasing skepticism. A serious supply chain attack has compromised 84 TanStack npm packages, prompting urgent calls for guardrails. OpenAI announced "Daybreak," a new cybersecurity initiative, while several open-source releases and tool integrations continue to accelerate the AI development ecosystem.
Key Events
-
Local AI reaches consumer-tier autonomy: Multiple posts demonstrate Qwen 3.6 27B dense running on single RTX 3090 (24GB VRAM) autonomously executing code, running tests, and handling workflows via Hermes Agent, challenging claims that cloud APIs are necessary for serious work. → tweet
-
Hermes Agent ships computer control capability: Nous Research releases native computer use functionality for Hermes Agent, allowing any model to control the user's actual computer while maintaining full user control of keyboard, mouse, and screen. → tweet
-
Major npm supply chain attack ongoing: 84 TanStack npm packages compromised in "Mini Shai-Hulud" attack; security researchers urge developers to stop all work and implement guardrails; RubyGems also reportedly under malicious attack. → tweet
-
OpenAI launches Daybreak cybersecurity initiative: Sam Altman announces Daybreak, an effort to provide frontier AI capabilities for defensive cybersecurity, acknowledging AI is becoming "super good" at cyber defense. → tweet
-
Nous Research + HuggingFace integration: Hermes Agent added to HuggingFace local apps with native traces viewer, making agent development more accessible. → tweet
-
Cerebras IPO speculation: Reports indicate Cerebras expected to go public at $35B valuation; @TheAhmadOsman advises against investing. → tweet
Analysis
Local AI Acceleration: A clear pattern emerges of users validating that open-source models (particularly Qwen 3.6 27B) running on consumer GPUs can now handle production agentic workflows previously requiring cloud APIs. The discourse has shifted from "if local AI can compete" to "how to implement it," with Hermes Agent positioning itself as the definitive harness for this use case.
Framework Consolidation: OpenClaw is facing significant backlash as users migrate to Hermes Agent, citing better performance with the same underlying models. This suggests the differentiation in the agent framework space lies not in the models but in the orchestration and tool-calling infrastructure.
Security Escalation: The npm supply chain attack represents a significant escalation in software supply chain threats, extending beyond a single package registry. This will likely accelerate adoption of code review tools like Codex's hook review feature and drive discussion around dependency verification.
AI Productivity Tools Maturing: Voice-based prompting, session pinning for cost tracking, and integrated accounting automation via AI agents indicate practical productivity applications moving beyond demos into daily workflows.
What to Watch: Monitor adoption metrics for Hermes Agent vs. OpenClaw, evolution of the npm supply chain attack and industry response, and whether local AI demonstrations translate to sustained enterprise adoption.
Tweet Feed
AI Models & Research
@TheAhmadOsman · 2026-05-11T22:15
Top 26 Essential Papers for Mastering LLMs and Transformers... [full list covering Attention, BERT, GPT-3, Scaling Laws, Chinchilla, LLaMA, FlashAttention, RAG, InstructGPT, DPO, CoT, ReAct, DeepSeek-R1, Qwen3, MoE architectures, and more] → tweet link
@Ex0byt · 2026-05-12T17:52
Pareto knee (zero loss) PRISM-DynamicQuant recipes for Qwen3.6-27B-PRISM-Pro are ready! Quants in nvfp4/gguf/mlx will be up on HuggingFace. → tweet link
@victormustar · 2026-05-12T15:38
RT @dlouapre: Meet physics-intern🧑🎓, our agentic framework for theoretical physics. It takes Gemini 3.1 Pro from 17.7% to 31.4% on CritP... → tweet link
@louszbd · 2026-05-12T15:03
great time on @tphuang's show. we got into glm and what China's AI scene feels like day to day. → tweet link
@jack · 2026-05-11T22:56
RT @miramurati: Today we're sharing our work on interaction models. A new class of model trained from scratch to handle real-time interactions... → tweet link
@sama · 2026-05-11T22:51
speaking of things that have gotten over a threshold for me, the combo of the new ChatGPT model, personality, and personalization feels like a new thing → tweet link
@sama · 2026-05-11T22:49
would you call it a superapp? → tweet link
@victormustar · 2026-05-12T10:28
RT @ash_csx: We're dropping two open source SLMs this week. 1. One of them matches SOTA accuracy at up to 93x smaller. 2. The other one... → tweet link
Local AI & Hardware
@sudoingX · 2026-05-12T08:16
look anon, those of you who kept saying local AI is not there yet... every pixel on this screen was written by one model, in one shot, on a single rtx 3090 with 24gb of vram. the model is qwen 3.6 27b dense q4. the harness is hermes agent. → tweet link
@sudoingX · 2026-05-12T13:32
my dgx spark just woke up and chose violence. step-3.5-flash-REAP-121B loaded. 121 billion parameters, 11B active per token, running locally on unified memory. It asked using hermes agent if i want to install dependencies and fire the test suite. i said yes from my phone while stuck in bangkok traffic. by the time i got home it was already executing code autonomously. → tweet link
@sudoingX · 2026-05-12T16:51
i hear it often. what's the point of local AI when cloud models are so far ahead. and yes, frontier closed-source models are ahead... if you handle client financials, do you want that data sitting in anthropic's next training set... some tasks stay between me and my machine. not everything has to be cloud-driven. → tweet link
@sudoingX · 2026-05-12T04:14
let me say this out loud here: there is absolutely zero reason to use openclaw in may 2026. a general agent exists. hermes agent does coding, video editing, marketing design, research, browser automation, terminal work. one tool, all under your roof. → tweet link
@sudoingX · 2026-05-12T05:53
hey anon, let me tell you this again because you only listen when repeated. qwen 3.6 27b dense q4 is the top king on a single gpu in the 24gb tier. if you own a 3090, 4090, or 5090 mobile, leave your work to this beast. → tweet link
@sudoingX · 2026-05-12T06:59
RT @Scobleizer: My team switched weeks ago. → tweet link
@TheAhmadOsman · 2026-05-12T18:55
Demos in the local AI space between heterogenous devices are misleading - Stuff like mixing MacBooks / Mac Studios with GPUs / DGX Sparks. I haven't seen a single mature software stack in that space yet. To talk about it with high certainty is quite dishonest IMHO. → tweet link
@TheAhmadOsman · 2026-05-12T16:12
FYI, I wouldn't invest / bet on Cerebras → tweet link
@tinygrad · 2026-05-12T18:22
There will be many winners and losers in the next 10 years while the stack for AI compute matures. If you want to succeed, focus on timeless numbers like FLOPS/$, FLOPS/W, GB/s/$ and GB/$. Any focus on applications will have a short shelf life. → tweet link
@LinusEkenstam · 2026-05-12T17:24
This is an interesting take on the mouse pointer. I suspect they have orchestrated something quite simple and elegant using vision behind the scenes here. Mouse + Voice + Snapshot of Screen = Action. → tweet link
@LinusEkenstam · 2026-05-12T09:03
Unitree building mechs, these are strong machines of the future. get ready for everything to change at a very rapid pace. → tweet link
AI Agents & Frameworks
@Teknium · 2026-05-12T18:37
RT @NousResearch: Today's Hermes Agent Jam starts in 90 minutes! Join the Nous team in our Discord to share what you're working on → tweet link
@Teknium · 2026-05-11T23:26
RT @itsolelehmann: The best model setups to run on Hermes (by price tier): 1. If you have infinite budget: Go with GPT 5.5 or Claude Opus... → tweet link
@Teknium · 2026-05-11T23:26
Keeping up with twitter notifs keeps getting harder as we scale more - Apologies if I've missed any bug reports of yours or etc. → tweet link
@Teknium · 2026-05-12T00:05
Tables finally fixed and perfect in Hermes Agent's TUIs lol → tweet link
@Teknium · 2026-05-11T23:47
RT @Saboo_Shubham_: This is so insane. Use native Hermes Agent /goal as the orchestrator and Codex /goal as the builder. → tweet link
@Teknium · 2026-05-11T20:23
RT @kevinrose: The number one project I want to try but haven't yet is @NousResearch Hermes. → tweet link
@Teknium · 2026-05-12T01:44
RT @OpenRouter: Docs from @NousResearch on how to set up Pareto Code in Hermes → tweet link
@Teknium · 2026-05-12T02:04
RT @Sigma_Browser: Hermes agent running in private browser on free local models. Private by design. Local. Open-source → tweet link
@Teknium · 2026-05-11T20:13
Give our early preview of Computer Use (with ANY model) a try today! Built into the latest Hermes Agent and powered by @trycua - opens the door to any model, not just the frontier models in special modes - to control your actual computer. → tweet link
@sudoingX · 2026-05-12T18:03
OpenClaw faces skepticism as users switch to Hermes Agent... users are walking from the framework i've been calling a babysitting trap. and the harness i've been recommending is shipping NEW capability while the competitor faces "skepticism."... the throne is still not crowded. your cognition deserves a better tool. → tweet link
@sudoingX · 2026-05-12T05:44
this is the universal openclaw experience and nobody who actually uses it daily disputes it. the entire openclaw market is just people too invested in the workflow to admit the harness is the bottleneck. hermes agent ships the same model with zero babysitting. start the loop, walk away, ship. → tweet link
@sudoingX · 2026-05-12T00:12
openclaw is terrible on local models btw. there is a reason i keep saying hermes agent anon. → tweet link
@sudoingX · 2026-05-12T04:45
if you don't have a 3090 to run qwen 3.6 dense locally, take this. qwen 3.6 plus is alibaba's frontier tier flagship, free through nous portal right now for a limited time. → tweet link
@sudoingX · 2026-05-12T11:50
hermes agent from bangkok traffic. the agent moves faster than i do. → tweet link
@victormustar · 2026-05-12T11:31
RT @NousResearch: Thanks to the @huggingface team for adding Hermes Agent to local apps and shipping a native Hermes traces viewer! → tweet link
Developer Tools & Coding
@kunchenguid · 2026-05-12T18:37
time to reveal my HTML workflow with agents. HTML is the new markdown. Lavish is the new editor for your HTML artifacts. just tell your agent - discuss the technical plan with me using
npx lavish-axi. 100% open source and runs locally. → tweet link
@kunchenguid · 2026-05-12T01:29
when i first saw this "HTML is the new markdown" thing, i quickly dismissed it as "that's going to waste so many tokens" - i was wrong. i just tried it, and it rocks → tweet link
@levelsio · 2026-05-12T11:14
By far THE most annoying part of running a business for me is collecting receipts for my accountant... I made a Xero API key, gave it to Claude Code, and asked it to login and figure stuff out, then it just asks me which expenses still need a receipt... Much easier! → tweet link
@nummanali · 2026-05-12T13:24
Very impressed by Codex on requesting review on newly detected hooks... I installed hooks to set up @EntireHQ and Codex rightfully asked me to review them before running. Definitely should be the norm given the recent attacks happening. → tweet link
@nummanali · 2026-05-12T09:57
npx devrage - how much do you swear at AI. Me? 4140 times ^_^. Claude is the most sworn at, more than 2x that to Codex. → tweet link
@badlogicgames · 2026-05-12T15:05
looked into /goal in both cofex and claude codr and all i'm seeing are inferior versions of autoresearch. what am i missing? available from the @ShopifyDevs folks as a pi extension. → tweet link
@badlogicgames · 2026-05-12T05:49
RT @gonizahavy: Hmm, could not handle the FOMO of @antirez DS4 so I made it work on my Strix Halo using ROCm HIPify 🫢 → tweet link
@badlogicgames · 2026-05-12T09:28
just did a live demo showing how easy it os to build /btw in pi. totally botched the live demo because i don't know how to prompt anymore. skill issues. → tweet link
@thdxr · 2026-05-12T16:39
it's very important our inference business has customers world wide. the entire game is keeping GPUs busy. hard to do that if your customers are all in one timezone. → tweet link
@thdxr · 2026-05-12T05:56
since compaction is good now i've been keeping a session pinned per PR im working on and reusing it until it merges. what's nice about this is i can see the cost of the session and understand what the feature cost me. → tweet link
@thdxr · 2026-05-12T03:50
some of you have asked for more examples of how i voice prompt - here's an example - you can see how it's rambling, not very coherent and yet it has 10x more information than what i would be willing to type out. → tweet link
@thdxr · 2026-05-12T13:07
root issue is most employees aren't in properly motivating environments. it's adversarial where the employer and employee are seeing how much they can get over on the other. it's exactly why ai isn't going to magically make teams more productive. → tweet link
@steipete · 2026-05-12T07:03
RT @thsottiaux: Now that the Codex app is close to being the super app. What should the super duper app do? → tweet link
@steipete · 2026-05-12T07:02
Crabbox 0.12.0 is live. 🪟 Azure Windows desktop + WSL2, 🏠 Proxmox + Tensorlake providers, 🧰 preflight, failure bundles, phase timing, 🚑 keep failed boxes around for SSH debugging. Remote test boxes got much less slippery. → tweet link
@gdb · 2026-05-12T05:01
Having an agent in your meeting is such a futuristic experience: → tweet link
@FrameworkPuter · 2026-05-12T15:07
RT @badlogicgames: thanks @FrameworkPuter for this wonderful gift. i'll report back once i've set it all up and bastardized @antirez 's ds4... → tweet link
@FrameworkPuter · 2026-05-12T18:21
We have our ~monthly set of updates to respond to changes in silicon costs. DDR5 memory costs have been more stable over the last couple of months, while storage costs continue to increase. → tweet link
@badlogicgames · 2026-05-12T15:05
RT @uavster: Si estás en Madrid, pásate a escuchar a Ignacio hablar sobre IA, robótica y exploración espacial. → tweet link
Security & Supply Chain
@nummanali · 2026-05-12T12:45
Very serious now. Stop all work and focus on placing guardrails. It's bleeding outside of being npm only. → tweet link
@jezell · 2026-05-12T18:04
RT @hetmehtaa: List of Compromised Packages of Active NPM Supply Chain Attack - Mini Shai-Hulud (May 11, 2026) @tanstack/react-router 1.16... → tweet link
@RealGeneKim · 2026-05-12T14:31
RT @maciejmensfeld: We're dealing with a major malicious attack on @rubygems right now. Signups are paused for the time being. Hundreds of... → tweet link
@RealGeneKim · 2026-05-12T14:31
RT @ryancarson: 🚨 There's a major attack going on via npm right now. Do not install any packages right now. Talk to your agent ASAP and... → tweet link
@nummanali · 2026-05-11T21:56
RT @SocketSecurity: 🚨 BREAKING: 84 TanStack npm packages were compromised in an ongoing Mini Shai-Hulud supply chain attack, adding suspect... → tweet link
@thdxr · 2026-05-12T18:05
everyone's ideas for fixing the npm security issue shows how basically no one is capable of thinking at the scale of this problem. → tweet link
@badlogicgames · 2026-05-11T23:21
pi depends on this package, but not that version. i'm sure we'll eventually get fucked by this too, by proxy. i hate the npm model of dependency resolution. → tweet link
@steipete · 2026-05-12T17:14
RT @ZackKorman: Mythos found a single vulnerability in cURL (along with three false positives, and one issue they classified as a bug). → tweet link
@jack · 2026-05-12T17:55
RT @ProjectLoupe: loupe [NOUN] 1) A small magnifying tool used by jewelers to detect imperfections in gemstones. 2) An AI-powered vulner… → tweet link
Open Source & Tools
@victormustar · 2026-05-12T14:33
RT @populartourist: One of the most important llama.cpp PRs is still waiting for mainline: MTP support means mode… → tweet link
@badlogicgames · 2026-05-12T08:30
RT @antirez: Breaking! @ggerganov himself confirming that DS4 2 bit quants actually work very well! He is running AIME2025... → tweet link
@jezell · 2026-05-12T05:24
RT @davidpwalter: 👾 @godotengine WebGPU Beta is out now. 🚀5x faster performance, 🤖 Compute Shader Support, 📲 Godot's Mobile Renderer → tweet link
@jezell · 2026-05-12T05:21
Lance 6.0 released. Yippee. → tweet link
@RydMike · 2026-05-12T06:10
I know Impeller is sacred, but not having features like this and more, in the home grown Impeller renderer choice, is and will be very limiting for Flutter UI development. Refusing to change direction is worse, a potential downfall path. → tweet link
@nummanali · 2026-05-11T19:36
A unified MD experience like HTML → tweet link
@gospaceport · 2026-05-12T17:51
Was about to open a repo on GH and I thought...ya know what...maybe I just use gitea since it's mainly for local use and that feels like a good choice today. → tweet link
AI & Cybersecurity
@sama · 2026-05-11T21:34
OpenAI is launching Daybreak, our effort to accelerate cyber defense and continuously secure software. AI is already good and about to get super good at cybersecurity; we'd like to start working with as many companies as possible now to help them continuously secure themselves. → tweet link
@gdb · 2026-05-11T21:00
Daybreak: our umbrella effort for defensive acceleration, equipping cyber defenders with the best possible frontier AI capabilities. → tweet link
@jezell · 2026-05-11T22:21
Very cool. Can't wait to see what @juberti and co ship, because certainly OpenAI has to be working on similar problems internally. This type of approach is definitely the future of things like the Realtime API. → tweet link
Industry & Business
@levelsio · 2026-05-12T12:09
Wavespeed for GPU (Photo AI + Interior AI), Cloudflare for R2 storage and domain renewal, xAI for LLM AI API for all my sites, Backblaze for backups, Hetzner for VPS, Scrapingbee for scraping... → tweet link
@levelsio · 2026-05-12T18:38
Do you talk to AI more than your gf/bf/wife/husband? → tweet link
@levelsio · 2026-05-11T20:51
🏆 The Vibe Jam of 2026 sponsored by @cursor_ai + @boltdotnew + @heyglif + @tripoai is now being judged! → tweet link
@sudoingX · 2026-05-12T18:20
anthropic will ship hardware products faster than openai btw → tweet link
@LinusEkenstam · 2026-05-12T12:41
This is what adobe should have shipped like two years ago… → tweet link
@LinusEkenstam · 2026-05-12T11:32
This video goes hard. I love how Lovable is embracing swedish taste skills. → tweet link
@TrungTPhan · 2026-05-12T15:32
Sanjay Mehrotra legendary LinkedIn work experience: co-invents tech for NAND flash memory, co-founds SanDisk and runs for 28 years, sells SanDisk to Western Digital for $19B in 2016, becomes Micron Technology in CEO in 2017 and market cap since grown 29x from $30B to $860B → tweet link
@jezell · 2026-05-11T20:34
RT @theinformation: Cerebras is expected to go public at a $35 billion valuation, testing how much demand Wall Street has for AI chip start… → tweet link
@kunchenguid · 2026-05-12T07:53
what thinking machines cooked here is such a rare combination of a great product vision, truly innovative technology and the most nerdy and entertaining demos. i will be cheering for this company. hope they'll bring us something unique and amazing soon! → tweet link