H(ai)pe?
Hype or no hype — what's the deal with AI? Is it going anywhere? As things stand, not with those American models. The question is whether we even need them.
Claude Fable is BANNED outside the US. What we’ve been saying for ages (a year and a half) — that you can no longer build on it — has now become reality. For anyone who still didn’t want to believe it, it’s now fact. That’s a good thing, just in time as it were. Before you know it, you’re dependent on it, and anyone who now has to work with the top model Claude is out of luck. Yes, it’s true: the American models are the best and most advanced in many respects — only you often don’t need that at all. Then there’s the cost factor. OpenAI and Anthropic want to go public, and loss-making companies aren’t usually welcomed with open arms. So the price had to go up — considerably. Even saying hello to Claude Fable has become an expensive joke, and that’s no joke. You’re already paying $10/$50 for input/output, and anyone who’s ever developed software with AI knows how quickly those tokens add up (sorry, phone ticks).
This brings us to the question: do you need it? Or better still: do you want it? If you look here: https://isaiprofitable.com and then consider who’ll be paying these debts shortly, we’d venture to say it’s the users, ultimately. And that means: American AI is and remains staggeringly expensive, because those extraordinarily long red bars need to be paid back eventually.
Hermes Agent
There’s more under the sun, especially if you look towards open source, where some truly cool things are coming from: Open Code, Open Design, and Hermes Agent. And that’s what we actually wanted to talk about. Hermes is an “agentic harness.” This means it’s a piece of software with which you can build the AI you need — one that does the things you want, in the way you want. That same quality is also Hermes’s pitfall. Hermes isn’t the easiest app to set up, because an enormous amount is involved depending on what you want it to do. On Reddit forums you can see many people starting enthusiastically, getting something running, and then wanting to know what you can actually do with it — a funny phenomenon.
We use it for hardware research, as a RAG. Every morning it searches for important news relevant to Ikbenstil Computers — via desktop but also via smartphone. We can even take photos and have the AI figure out what it knows about the component pictured. Hermes’s strength is its memory. Anyone who’s worked with AI for a few days knows: on Monday the AI no longer knows what you did on Friday. Not so with Hermes: here you have the option of installing both short-term and long-term memory.
The problem with Hermes is that as a platform it quickly needs to communicate with a great many other apps. So there are quite a few API connections, databases (yes, plural), external AI connections and local connections involved. We use, for example, an API AI via Cortecs and a local image recognition AI.
But once again: it’s a platform, so you can make of it what you will — or as many Reddit users say: going down the rabbithole. But impressively cool. This is AI that actually does things and is capable. Small warning: yes, you can connect Hermes to just about anything — email, databases, image creation, audio recordings… you can’t think of it wild enough. Meaning: the best route is to experiment, get to know it, and only once it’s “under control” connect it to apps that matter.
And — since we started on the topic of American AI models — Hermes can be used entirely locally, and it works excellently. Yes, that means complete, 100% privacy. You just need enough VRAM. Something there isn’t exactly an abundance of at the moment. But you do have the advantage: new model, new capabilities = use model, done. Our tip: use models that can handle tools and skills.
One more tip: run it locally on dedicated hardware. If you use API LLMs, a Raspberry Pi is already sufficient. Docker is cool but can also get in the way. There are people who install this on a VPS — that works, but it adds a layer of complexity.
Here’s the capability list:
1. Core Identity
- Open-source AI agent framework by Nous Research (the lab behind Hermes, Nomos, Psyche models)
- Self-improving — the only agent with a built-in learning loop: creates skills from experience, improves them during use, persists knowledge across sessions
- Provider-agnostic — works with 20+ LLM providers (OpenRouter, Anthropic, OpenAI, DeepSeek, Google Gemini, xAI, local models via LM Studio, custom endpoints, and more)
- Runs anywhere — Linux, macOS, WSL2, Windows (native), Android (Termux); on a $5 VPS, a GPU cluster, or serverless (Daytona, Modal)
The full capability list (60+ tools, multi-agent orchestration, memory systems, security, deployment options and 20+ supported LLM providers) can be found in the original article — the technical specifications have been intentionally kept in English.
Source: Hermes Agent documentation (hermes-agent.nousresearch.com), skill database, and live system inspection. June 2026.