← Tech / AI / IT Monitor Index Tech / AI Generated 2026-06-28 19:31 UTC

Tech / AI / IT Monitor

June 28, 2026 · Based on tweets from the last 24 hours · 124 tweets analyzed · model: ollama-cloud/glm-5.1:cloud

Executive Summary

The last 24 hours saw a major escalation in the open-source vs. closed-source AI debate, driven by Anthropic's Dario Amodei making comments perceived as dismissive of open-source, sparking a fierce backlash from the local AI community. Meanwhile, Exo announced a strategic partnership with NVIDIA to make local AI "the default," with major reveals planned for the Local AI Summit on July 2nd. The Hermes Agent ecosystem from NousResearch continued to mature rapidly, demonstrating Mixture-of-Agents (MoA) capabilities that reportedly beat Opus 4.8 by 8% and GPT 5.5 by 11% on benchmarks. On the hardware front, comprehensive comparisons of memory bandwidth across local AI platforms highlighted the emerging competitive landscape between NVIDIA, Apple Silicon, AMD Strix Halo, and Tenstorrent.

Key Events

Analysis

Patterns observed: A clear fault line is deepening between the local/open-source AI community and closed-source incumbents. The Fable 5 access revocation (referenced multiple times by @sudoingX) has become a rallying point—demonstrating that API-dependent users can lose access overnight, while local model owners retain full control. This is accelerating hardware investment and local inference tooling adoption.

Escalation trends: The community is rapidly consolidating around a "full local stack" vision—Hermes Agent + local models + personal hardware—as a coherent alternative to renting cloud APIs. NVIDIA's willingness to partner with Exo on local AI signals institutional recognition of this movement. Meanwhile, the quantization and hardware bandwidth discussions reveal an increasingly sophisticated user base that understands local inference at a deep engineering level.

De-escalation signals: Some voices (e.g., @hnasr on software bloat, @thdxr on agent frameworks) are pushing back on hype cycles, calling for more substance over framework proliferation.

What to watch next: The Local AI Summit on July 2nd is the key near-term event—expect concrete NVIDIA + Exo deliverables. Watch for whether Hermes Agent's MoA benchmarks hold up under broader testing. The open-source vs. closed-source rhetoric could further escalate if Anthropic or OpenAI announce new model restrictions. On hardware, watch AMD Strix Halo and Tenstorrent as emerging x86/fully-OSS local inference alternatives.

Tweet Feed

Local AI / Open-Source Movement

@alexocheema · 2026-06-28T18:06

We've been working with NVIDIA in their HQ for the past month. We're going to make Local AI The Default. BIG news to share at Local AI summit, SF, July 2nd. → tweet

@TheAhmadOsman · 2026-06-28T17:01

MASSIVE NEWS. Teamed up with NVIDIA to make Local AI The Default → tweet

@TheAhmadOsman · 2026-06-28T03:53

New bio as we enter a new era in which Opensource AI WINS. → tweet

@TheAhmadOsman · 2026-06-28T01:34

Cancel your Anthropic and OpenAI subscriptions. Hurt them in their pockets and watch them crawl back begging for our money → tweet

@TheAhmadOsman · 2026-06-27T23:27

To folks at Anthropic / OpenAI / etc: History will judge us harshly, and deservedly so. Ask yourself honestly, are you building an open and free world for your kids or just cashing a bigger check? → tweet

@TheAhmadOsman · 2026-06-28T04:58

Moving forward, I will tell nobody that Opensource models are as capable as Claude. We are always behind their dangerous models, apply all your safety rules to them tyvm → tweet

@TheAhmadOsman · 2026-06-28T03:25

Dario has a fearmongering machine. This narrative needs to be reversed ASAP → tweet

@Teknium · 2026-06-28T05:18

Uhh dario you realize that Linux is largely run on the cloud too, right → tweet

@Ex0byt · 2026-06-28T15:45

ouch... the anti anthropic sentiment on x is palpable today (just leave oss alone) → tweet

@juliarturc · 2026-06-28T02:33

The sentiment in the comments is so negative, this can't be good news for the IPO. Even if they did play 5D chess and summon regulatory capture just to project their tech is dangerously capable, it seems like it backfired. → tweet

@sudoingX · 2026-06-27T20:44

weirdly grateful tonight. the fable 5 ban reminded me why i went local in the first place. everyone else is panicking about losing access, and i'm just sitting here filling another nvme. my model collection has never been bigger → tweet

@sudoingX · 2026-06-27T21:19

you have no idea what i put my local models through every day. i crack them fucking open man, tear the weights apart, tune them on whatever i want, push them till they break, then rebuild them how i like. this is the feeling nobody renting an api will ever have. → tweet

@levelsio · 2026-06-27T19:30

It took 2 years for my "come and take it GPU" design to get noticed but now they are as the US government is clamping down on AI model access → tweet

@steipete · 2026-06-28T02:50

"History teaches us that access blockage rarely stops determined users." → tweet

@sudoingX · 2026-06-28T10:54

คนไทยนอนหลับทับ local AI กันอยู่รึเปล่า อเมริกาเพิ่งดึง Fable 5 ออกทั้งโลกในคืนเดียว ถ้ายังเช่าโมเดลปิดอยู่ คุณไม่ได้เป็นเจ้าของอะไรเลย ตื่นได้แล้ว มารันเอง → tweet

Hermes Agent / NousResearch

@sudoingX · 2026-06-28T17:23

spent the last hour just enjoying this, and i have to say it: the Hermes Agent dashboard is beautiful. a real agentic stack, near lossless, with a UI this clean, running entirely local on a single DGX Spark. → tweet

@Teknium · 2026-06-28T01:22

RT @HermesAgentTips: your hermes agent can now run 3 frontier models on the same query and have claude opus synthesize the best answer → tweet

@Teknium · 2026-06-28T11:13

RT: Hermes Agent's MoA beats Opus 4.8 by 8%, GPT 5.5 by 11% on HermesBench → tweet

@Teknium · 2026-06-28T18:49

RT @DavidOndrej1: I gave my Hermes Agent a phone number. now it makes calls, answers calls, runs tasks while I sleep. → tweet

@Teknium · 2026-06-28T05:37

RT: Hermes just turned "only for the chosen few" into "anyone with a signal and a stack." Frontier is officially crowdsourced now. → tweet

@Teknium · 2026-06-27T21:07

RT: Hermes Hackathon Submission: Allmind is an all in one business creation and management platform built from the ground up... → tweet

@sudoingX · 2026-06-28T15:43

you can auth x premium+ straight into hermes agent. that puts grok build, a frontier model, inside your own open agent, off the sub you already pay for. → tweet

Hardware & Quantization

@TheAhmadOsman · 2026-06-28T00:50

Local AI hardware = capacity X bandwidth X software stack. Hardware by Memory Bandwidth: [detailed comparison of Mac Studio M3 Ultra, RTX PRO 6000, RTX 5090, DGX Spark, Strix Halo, Tenstorrent, etc.] The only mental model that matters: 1. What must fit? 2. What bandwidth tier do I need? 3. What software stack can actually deliver it? → tweet

@sudoingX · 2026-06-28T18:40

here is the quant cheat sheet nobody gives you straight. save this if you run local models: Q4_K_M = smallest, fastest, real quality loss. Q5/Q6 = middle ground. Q8_0/FP8 = near lossless, the sweet spot. bf16/fp16 = full precision, the quality ceiling. → tweet

@sudoingX · 2026-06-28T18:53

okay nerds, how much memory do you actually own right now? not rented, owned. [Lists 448GB across DGX Spark, Strix Halo, 5090 laptop, 3090 node, 3060 node, etc.] → tweet

@cooltechtipz · 2026-06-28T03:47

Liquid cooling for high-density AI systems. → tweet

@FrameworkPuter · 2026-06-28T01:55

SteamOS 3.8 runs great on Framework Desktop. You can drop into desktop mode to use it as a general-purpose workstation too. → tweet

@FrameworkPuter · 2026-06-28T17:31

There are levels to clustering… → tweet

AI Models & Inference

@ollama · 2026-06-27T20:46

Run Ornith with Ollama: ollama run ornith. For coding, use it with Claude or Pi. For the more capable 35B model, use: ollama launch claude --model ornith:35b → tweet

@Ex0byt · 2026-06-28T14:33

Curious is Grok 4.5 free of older Cursor Composer RL Data which started from open-source base (Kimi K2.5) + heavy fine-tuning/RL? → tweet

@Ex0byt · 2026-06-28T17:51

when will we have a GLM-5.2 capable version of Grok? → tweet

@TrungTPhan · 2026-06-28T17:53

RT: Bloomberg chart showing amount of RAM needed for AI data centres. Integrated server rack of 72 Nvidia Blackwell chips = same... → tweet

@sudoingX · 2026-06-28T17:48

dear anyone at xai... one thing keeps biting me: usage transparency. i hit the limit unexpectedly, no warning, no meter. i'm asking to SEE where i stand. a usage meter, a heads up before the cutoff. → tweet

@FinansowyUmysl · 2026-06-28T06:28

Bardzo ciekawy case study jak Coinbase zmniejszyło rachunki za AI, prawie o połowę, nie zmniejszając zużycia tokenów. [Coinbase AI cost reduction case study] → tweet

Developer Tools & Workflows

@levelsio · 2026-06-28T09:22

I think I've been coding almost solely on my VPS with Claude Code for almost a year now. It just keeps going all night while you sleep. It just feels like living in the future. → tweet

@thdxr · 2026-06-27T23:07

in OpenCode v2 all instances of the tui and desktop and web share the same backend. so everything is synced by default and resource usage is minimized no matter how many windows you have open → tweet

@thdxr · 2026-06-28T01:42

i need fable back so i can continue to build my minecraft clone but a little different → tweet

@thdxr · 2026-06-27T21:22

everyone's happy to keep building agent frameworks while ignoring every single agent in the products they use → tweet

@levelsio · 2026-06-28T10:44

If you use vanilla PHP and vanilla JS there is nothing to be restarted! → tweet

@levelsio · 2026-06-28T12:10

Made a Nullsoft SHOUTcast server. Winamp's own streaming platform that let you create self-hosted radio stations. Around 2011 it reached its peak with 900,000 concurrent listeners with about 45,000 active radio stations. But the tech is still there and you can host a server! → tweet

@victormustar · 2026-06-27T21:04

RT @GithubProjects: Chat UI is a SvelteKit chat interface that works with any OpenAI-compatible API, powering HuggingChat → tweet

@ASalvadorini · 2026-06-28T09:17

Lately a mate and I, we are prototyping the feature while in the meeting where the managers and the team discuss the feature. By the end of the meeting we show them a half baked feature fully working. #AI → tweet

@jezell · 2026-06-27T20:59

RT: All you need for AI type design is img2bez and Runebender-WebGPU in Codex. Here Codex is using the OpenAI image API and img2bez... → tweet

Software Bloat & Performance

@hnasr · 2026-06-28T14:01

We used to hear floppy disks and even hard drives every time an application reads and writes. IO was expensive and you could feel it. This forced devs to build efficient applications. As IO got cheaper, software got sloppy... Modern apps got so slow that an app from the early 2000s ran on ancient hardware is somehow faster. Now we have LLMs trained on this modern software, which inevitably will produce more software that demand more hardware. Performance and troubleshooting engineers who understand the atomic fundamentals of software will be in high demand. → tweet

AI Industry & Strategy

@cooltechtipz · 2026-06-28T16:55

Over the past decade, several technologies have evolved from commercial tools into strategic national priorities. [Lists nuclear, GPS, semiconductors, frontier AI models.] Frontier models are increasingly being treated more like advanced chips or cybersecurity, where innovation is shaped by both market demand and national security. → tweet

@jezell · 2026-06-27T21:09

RT: The top 1% of U.S. AI firms are now spending about $7,500 per employee each month on AI. → tweet

@TheAhmadOsman · 2026-06-27T20:22

Wanna replace Anthropic/OpenAI? START WITH THIS. The bible for running LLMs locally is now available online to read for free. Covers what to use on laptops, Macs, single GPUs, multi-GPU, production serving, and cluster orchestration. → tweet

Ambitions & Future Roadmaps

@sudoingX · 2026-06-28T10:17

a new era of compute is coming anon, and a whole new wave of local AI users with it. the market right now is scattered, there is different tool for every problem... i can see exactly what that looks like. → tweet

@sudoingX · 2026-06-28T10:01

i'm not building for today's user. i'm building for the one who shows up in 2028 and assumes it was always this easy. nobody will be hunched over a laptop wiring up environments by hand. it'll all be agent managed, one window over any environment, anywhere. → tweet

@sudoingX · 2026-06-28T11:10

my name is sudo and i'm 26. i am going to build the biggest data centers in southeast asia. not to go chasing users. because the demand from what i'm building will get so big i'll have no choice but to own the metal myself. → tweet