Recently I started using an AI Agent tool called Hermes, you may have heard of it. In fact, I think you do, at least a little, because you clicked on an article just labeled "Hermes" on a site called "The IT Journal". I don't think you would be looking for a reference to "Hermes" the Olympian deity in ancient Greek mythology. I guess maybe you were looking for the watch company, but again — watches, IT, journals — I guess they go together to an extent, idk. Anyway, I guess for the uninitiated I should tell you a bit about what Hermes is, or even what an AI agent is... actually, no, you can find a gazillion videos on what those are. Here is what Hermes is.
Hermes is a tool, a harness for AI but unlike Openclaw, its methodology is better placed, it was not a weekend project, not even a response to Openclaw. In fact, Hermes has been around for longer than Openclaw, built by a company called Nous Research. Nous built this for helping test agents, and other stuff, and the story goes that when they saw Openclaw come out and used it, they thought it was clunky, yada yada. They published it, and as a bunch of people saw how much more polished it was than Openclaw, people started using it. I don't know if it was Networkchuck who helped it get to even more people, but I found it through him. I had never used Openclaw, mainly because I like programing and had built a bunch of stuff from scratch, stuff for talking via Telegram, making cron jobs, etc... and since all I used it for was server reports, security response helper, and stuff like that, I felt that Openclaw was too big. Using npm was not something I liked or like, due to supply chain attacks, as well as it never making sense of why people needed so many packages (we could talk about if most are really needed or just bloat a codebase and fail to add any real benefit). When I first saw Hermes I was thinking the same thing, why do I need/want this. And the fact is... that I really didn't. But it's cool, and over the past week or so (sike, idk how long I have taken to write this), I have liked the memory, the more nuanced system prompts that I don't need to write, the ever-changing nature of my agent (Scot), as well as the ease of changing stuff (it changing stuff). I do not have many guard rails yet (I will implement more when it breaks something and I need to :), and since LLMs are getting really good, and really cheap, I feel fine letting it have a small server instance of its own.
It's really useful for making quick changes while out walking, it's not like FaceTiming a person and saying hey can you do this, but it is like just a text away. I still don't know how I feel about it not always asking for permission to run a command like my older system did/does, but it's got limited (root) access to the server it's on... I hope it lives :), I have been very pleased with its capabilities and how much it has helped with server setup. It's still a WiP, but all in all I like it. I am still finding out what's the best models for what, at this point in time. Models are coming out left and right all the time, so that will forever be changing. I may make a cron job that runs weekly that finds the stats of models on Ollama cloud and picks the best model for the different jobs I have (IT setup, Security, General cheap chat, and Planning), but really with just 2 main models right now, Minimax M3, and GPT 5.4, all is working nicely.
I have come to the conclusion during the writing of this article, that this was a ramble. There is just about no structure, no distinction of ideas, and just about no helpful info. But it does offer a view into how somebody (me) uses Hermes, how I found it useful, and maybe make you go watch Networkchuck's video about it... who knows? Agents are really good tools. They help in the small ways that give you time to... In "@nopenope-j5s" words in a video about Fable 5, "I'm super glad AI is beating games for me so I have more time to clean dishes and work in the lithium mines."
Anyway, this post seems like a bad one. I shall post it, but that's because hardly anyone will read it, but don't worry people who do, the BAB'ing part 2 will come out in the nearer future and will be better than this :)
Bye.