Eine sehr gute Auswahl an Top-Plugins für Hermes
9 EXCELLENT HERMES PLUGINS. ALL WORTH A WEEKEND. 🔖
All covered on HermesWatcher this week. Pick one, set it up, see what your agent can do by Monday.
▸ Make it drive your phone: Phone Harness
▸ Make it run your home: Home Assistant
▸ Make it answer Siri: Hermes Shortcuts
▸ Make it research a spreadsheet: Hermes Sheets
▸ Make it work through a plan: Hermes Whiteboard
▸ Make it turn one video into a week of posts: Content Engine
▸ Make it sort your X bookmarks: X Bookmarks
▸ Make it a full image studio: ImageGen
▸ Make it play YouTube beside your chat: YouTube Dock
Full post and demo for each one below. 👇
1
In der Tabelle finden sich die vom Ökonom Soumitra Shukla zusammengedtellten Top 20 Aufgaben, die am wenigsten von KI gelöst werden oder erledigt werden können.
2
Ein sehr beeindruckendes OCR-Modell mit nur 3 Milliarden Parametern, 100 Seiten Dokument in einem Rutsch durcharbeiten
China open-sourced a peanut-sized OCR that parses entire 100-page PDFs in one shot..
It's called Unlimited-OCR. Only 3B params. Runs locally.
Every other OCR tool chops your doc into pages and loses the thread. this one reads the whole thing in a single pass.
→ One-shot "long-horizon" parsing (32K context window)
→ Multilingual, out of the box
→ 93% on OmniDocBench v1.5 (+6 over the DeepSeek-OCR baseline)
→ Edit distance stays under 0.11 past 40 pages
→ Runs 100% locally on your own hardware
→ Works with Transformers, vLLM, SGLang, and Docker
Traditional cloud OCR (Textract, Google Vision, Azure Document Intelligence) often costs about $1.50–$15 per 1,000 pages.
This runs on your machine. For free. Forever.
Baidu built it to push DeepSeek-OCR one step further. It now has millions of downloads on Hugging Face, and most people still have no idea it exists.
100% open source (MIT).
github.com/baidu/Unlimited-O…
3
sehr gutes Video-Generierungs-Tutorial, das zeigt, wie man eine Schicht-Animation erzeugt, zB ein Holzboot, dass sich schichtweise zusammensetzt
Bring your images to life one piece at a time with a sequential layer effect using Omni 1.2 Flash in Google Flow.
Here is the four-part prompt recipe to build the effect:
1. Reference your image directly within your prompt.
2. Prompt for a “design animation.”
3. Specify the primary element to focus, like the towering skyscraper in this example.
4. Clearly describe how each layer should be added sequentially, one by one.
What will you create? Tag us, we want to see!
3
Die Anwendungsbereiche von KI sind wirklich uneingeschränkt: Eine KI überwachte Bettwanzenfalle für Hotels. KI ist wirklich eine Allzwecktechnologie, die man in jedem Bereich anwenden kann.
hotels are now hiding AI traps under the mattress that catch bed bugs before you ever meet them
and guests went from finding 50% of infestations to just 3.3% in its first year
> the trap gives off a scent bed bugs love and humans can't smell
> when one crawls in, a camera snaps a photo and AI checks it's really a bed bug
> then housekeeping gets an alert with the room number to clear them out
so your next hotel bed might have a tiny AI bed bug bouncer living under it lol
8
Das scheint ein wirklivh sinnvolles Plugin zu sin, dass die Aufgaben auf passende Modelle verteilt, je nach Komplexität
Most Hermes users are burning premium usage on work a cheap model could handle.
Lookups, summaries, research passes. All running on your most expensive model.
Advanced Delegation fixes that. It's an official Nous Research plugin that lets every Hermes subagent run on its own model.
▸ A cheap or free model for the grunt work
▸ Your premium model only for the hard thinking
▸ A local model on your own endpoint if you want
It also ships with a model-routing skill. Hermes sends each task to the cheapest route that gets it right, and only moves up when a cheaper model fails.
Quick setup:
▸ hermes plugins install advanced-delegation
▸ hermes delegation config
▸ Pick Model routes, add a cheap route and a premium route
▸ Start a new chat
In Desktop it's Settings → Model → Subagents.
Same agent. A lot less usage burned.
Beeindruckendes SEO-Skill-Repository, was hier zur Verfügung gestellt wird
Open-sourced my full SEO and AI-search stack for Claude Code and Cursor (30 skills + 4 live MCP servers I actually use every day)
1. 30 open-source SEO skills (technical audits, programmatic SEO, BOFU pages, directory submissions, Ahrefs/GSC workflows):
github.com/Illyism/seo-skill…
2. Full walkthrough and chained workflows for how I run them:
il.ly/blog/seo-skills
3. Video and YouTube transcript MCP (pulls captions + frame OCR right into your agent):
scribiz.com/mcp
4. Live AI search visibility tracker MCP:
aiseotracker.com
5. Hands-on tool directory & benchmark MCP:
awesome.tools
Install all 30 skills in one line:
npx skills add Illyism/seo-skills
Most SEO agency retainers are basically 5 of these checklists run manually once a month tbh. Easier to just hand them to your coding agent and merge the PRs yourself!
2
87
Das sieht schon sehr beeindruckend aus. Dass der Beweis der Riemann-Vermutung 250.000 Seiten umfasst, der von der KI produziert wurde. Und dass ein menschlicher Prüfer circa 14 Jahre dafür braucht, um den Beweis durchzuprüfen
Aber allem Anschein nach, wenn man hier auch im Video sieht, wie sich hier Experten dazu äußern, scheint hier wirklich ein phänomenaler Durchbruch erreicht worden zu sein.
卧槽,看来Claude助力黎曼猜想证明成功这事是真的,刚刚刷到弦理论领域知名学者、哥伦比亚大学数学教授 Brian Greene 的最新视频,
信息量大到有点不真实,看完有种一脚踩进科幻小说里的恍惚感😭
我拖回去反复看了三遍才有点回过神来,他在视频里披露了一个足以炸穿科学界的消息,数学家 Levent Alpöge 宣称借助 Anthropic 的 Claude 攻克了黎曼猜想,甚至连带着把推广版本也一并拿下了,
而整份证明足足有 25 万页,目前地球上还没有任何一个人类真正读懂它,
我第一反应跟所有人一样,怀疑是不是 AI 又在一本正经地胡说八道凑字数,
但看到视频里披露的那个硬核细节之后,我整个人瞬间清醒了,这份 25 万页的证明全是用 Lean 做了形式化验证的,
写过代码或者搞过数学的朋友都清楚 Lean 有多冷酷,它不像普通论文可以靠模糊话术蒙混过关,每一行推导都必须过得了定理证明器的机器编译,只要中间有一丁点逻辑跳步或者幻觉,系统当场就会报错卡死,
在机器逻辑的底层尺度上,这 25 万页天书已经跑通了严丝合缝的自动化核验,
可最让人感到魔幻的地方也恰恰在这里,如果按一个人类数学家每天精读 50 页硬核论文的速度去啃,不吃不喝也要整整 14 年才能翻完,我们大脑的生物带宽,已经装不下硅基智能展开的推理路径了,
退一万步讲,哪怕后续数学界在形式化定义里挑出了瑕疵,哪怕这个消息最后打了对折,它暴露出来的那个技术拐点也已经无法逆转了,
过去几千年,数学真理必须先被人类理解才算成立,今天 AI 加上 Lean 形式化验证,第一次把真理与理解强行拆开了,
机器已经冷冰冰地确认了答案正确,可全地球最聪明的一批大脑站在那 25 万页推导面前,连第一遍通读都做不到,
而这事真正吓人的地方根本不在数学界,在即将被 AI 接管的整个软件和商业世界,
当未来 AI 每天能生成几十万行人类根本看不懂的复杂代码和金融策略时,
靠人类工程师一行行做代码审查已经行不通了,AI 生成推理加上形式化自动验证,很快就会变成所有企业级 Agent 落地的生死线,
人类再也当不成真理的发现者和代码的编写者了,往后我们更像站在硅基黑箱外面负责签字的验收员。
21
Das Modell kommt auf alle Fälle auf die zu Testen Liste
AMERICA IS BACK
Thinking Machines just launched Inkling 2 Small. A 26B model that fits on a laptop
O, and it mutilates Qwen 3.8 27B. Faster and smarter
Thinking Machines figured out how to fit significantly more intelligence in a significantly smaller blueprint
Unlimited superintelligence for all is now back on the table
Democratized access to all knowledge is now back on the table
Highly recommend going to your agent of choice (grok bot, Hermes, Dots, Duet) and giving them this post. Then tell them to download and run this model
Then be blown away by the fact you have unlimited superintelligence on your desk
5
Eine sehr gute verdichtete Beschreibung der in 2026 erreichten neuen Höhepunkte mit künstlicher Intelligenz, die wahrscheinlich im Rückblick einen großen Wendepunkt darstellen werden
History was made today. Humanity will look back on 2026 and think: this was the threshold we crossed, the moment the future changed for everyone, forever.
People probably first sensed that something was shifting in late 2025 with Claude Code, when agentic AI became truly tangible for them. In the first half of 2026, things picked up speed. Model releases became more frequent, and everyone could feel the gaps between models and the improvements becoming more and more tangible. In the second half of 2026, however, there was a qualitative leap. It was no longer just the models getting better. It was the insights, the results, the outputs, the scientific solutions produced with AI that became more impressive, and today they have probably reached their absolute peak.
OpenAI led the way and solved the Navier-Stokes problem, a Millennium Prize Problem that humans had failed to solve for a very, very long time. A few weeks later, they published results on more than 300 open mathematical problems, many of which had also gone unsolved for decades. The entire scientific world was stunned, frozen in shock. But today Anthropic went a step further. Levent Alpöge, working with Claude, has almost certainly proven the Riemann Hypothesis. This is, and it’s hard to put into words, incredible and unfathomable. Today, I would say, is the proof of superintelligence. One of the greatest mathematical problems, if not the greatest, unsolved for 167 years, the holy grail of mathematics and science, has fallen. And it follows, and this should be clear to everyone, that all other problems will fall too.
We know that the development of AI models keeps moving forward relentlessly, and that this is just one step on the path to ever better models. And if the Riemann Hypothesis has already been proven today, there is simply no telling what even better models will solve. Dario Amodei said that AI will in all likelihood cure diseases, perhaps even all diseases. And although these are two different scientific disciplines, the significance of these breakthroughs shows that this is becoming more and more likely. 2026 will go down in human history as a turning point, the moment the singularity became tangible. And 2027 will be the year the singularity breaks into reality with full force and full impact.
I am in shock and excitement and I don't know what to think. But deep inside if feel joy.
This day is one of the best days ever for all of us.
5
Super, dass europäische KI nun einen Desktop-Platz erhält. Erster Test der App zeigt ein solides Funktionsspektrum mit inapp-Browser, Ordner-Zugriff und Computer-Use. InApp-Browsing noch sehr beschränkt auf öffnen von Seiten aber ist auch gerade erst heute rausgekommen, da wird bestimmt noch viel ausgebaut
Mistral Vibe is now available on Desktop!
Run agents locally or remotely, and start shipping straight from your MacOS or Windows device.
You can try the Pro plan for free if you’re building in Europe. Coupon code: LECHONK
(yes, we added more after we ran out yesterday).
Use the link in 🧵:
5
Eleven Reader wurde für die Top-App des Jahres 2026 vorgeschlagen und auch aus meiner Sicht zu Recht, denn es ist eine einfache, aber extrem praktische App, sich sämtliche Inhalte und Dinge vortragen zu lassen. In optimaler und immer besser werdender Stimmqualität, und zusätzlich auch noch die Möglichkeit zu haben, mit den eingebrachten Inhalten in Dialog zu treten
17
Interessanter Hinweis, dass es nun von Google ein Hilfedokument gibt, bei dem beschrieben wird, wie es möglich ist, Google-Agenten passend Zugriff auf die Webseite und Inhalte zuzulassen
Google a publié un nouveau document d'aideexpliquant comment autoriser Google-Agent à accéder à votre site web. « Google-Agent est le nom utilisé par les agents hébergés sur l'infrastructure de Google pour naviguer sur le web et effectuer des actions à la demande des utilisateurs », précise le document.
developers.google.com/crawli…
7
KI hilft bei der Krebsfrüherkennung durch die Datenauswertung der Atemkontroll-Datensätze. Interessants Verfahren
i wish more AI doomers saw stories like this one
there's a wholesome startup that trains beagles to sniff your breath for cancer, then AI reads whether the dog has smelled it
and in a study of nearly 1,400 people, it caught breast, lung, colorectal and prostate cancer 94% of the time!
it works because cancer cells give off chemicals that leave a faint smell in your breath, and dogs have about 40x more scent receptors than we do
1. you breathe into a mask at home and mail it to the lab
2. a beagle sniffs it and sits down if it smells cancer
3. sensors on the dog also catch the smaller tells, like a heartbeat spike or a sniff that lasts too long
4. AI reads all of it and turns it into your result
and it caught 95% of cases at the earliest stages, which is when cancer is most treatable
incredible tech
5
Das interessant, Messung am Terminal-Bench Benchmark der zeigt, dass bei Modell-Einsatz durch mehere Agenten die Ergebnisse deutlich besser sind und es bei den Modellen auch noch deutlich unterschiede gibt
Replying to @perplexity_ai
On Terminal-Bench 4.1, we ran five frontier models alone, as two and four copies, and leading a mixed-model team of four.
In Multiagent, each mixed-model team outperformed both the lead model and multiple copies.
The mixed team led by Gemini 4 Carbon scored highest, at 0.727.
4
Schöne Anleitung: Wie baue ich mit KI meine eigene Schrift, die ich später als Computerschrift verwenden kann?
Turn a rough sketch into a working custom font in minutes with Gemini 4 Hydrogen.
Using Google @Antigravity in the Gemini app for Mac, Hydrogen handles the technical heavy lifting so you can build custom typefaces and interactive design tools from scratch.
1
8
Google Flow teilt hier mit den Beispielen quasi einen Spickzettel zu Kamerabrennweiten. Das hilft, bessere Prompts für die eigene KI-Bildgenerierung zu erstellen, inklusive Effekten wie Schärfentiefe und Verzerrung. Funktioniert nicht nur in Google Flow, sondern auch mit anderen KI-Bildgeneratoren - siehe meine Beispiel hier, erstellt mit ChatGPT Image:
Save this camera lens cheat sheet for your next generation.
Which lens focal length do you use most often in your prompts? Drop your favorite in the comments.
And remember, image generation costs 0 credits in Google Flow. Try it out!
1
17
wird das ein Zukunftszenario sein, dass die überwiegende Teilnehmerschaft bei Webmeetings aus KI-Agenten besteht?
Joke aside: Why should we still hold meetings in the age of AI agents? My take: within a year, we will replace human meetings with team meetings involving AI agents.
Just as my own "Grokbots" already hold team meetings twice a day, which I simply attend mostly as an observer.
1
das ist beindruckend, dass dieser Film komplett mit Code erstellt wurde. Repository gibt es hier: github.com/BubbleBrainBB/awe…
🚨 When Fable 5.2 give this video to me, I was genuinely blown away.
Everything you see, and even everything you hear, was created entirely with code.
Now, let Fable 5.2 take you on a journey through the world as seen through the eyes of a little cat.
17































