Hermes Agent

“I made a sketchnote from my first month using Hermes Agent. My initial assumption was that it was basically another AI coding assistant / Claude Code alternative. After using it for a while, I think that framing is too narrow. The interesting part is not the chat box. It is the system you can build around it. The parts that changed the experience for me: SOUL.md turns it from a polite customer-service bot into something closer to a working teammate. Memory makes it stop “meeting you for the firs”

“A mi experiencia por mucho usar modelos como GPT 5.5 o Minimax-M3 es sumamente estupido. Hay que repetir demsaiadas cosas y es un sistema demasiado rebuscado. A la larga lo que he acabado haciendo es que Claude Code/Codex fabrique las cosas que necesito. Como standups matutinos, recordatorios de trabajo, investigacion de RRSS, etc.. Hermes puede estar bien para non-tech users pero se queda corto cuando pides muchas cosas o requiere tantas customizaciones que es lo mismo que programarlas tu mismo”

“Router-first tool loading + memory tiering — how it works and what it actually saves Been running this for a few days and the results were significant enough to share. This post covers the architecture, the numbers from my setup, and a quality regression to show it doesn't break anything. Public reference repo at the end. The problem Before any of this, every first-turn prompt in Hermes carried the full tool schema payload every toolset, every definition, regardless of what you asked. Typing "hi”

“You want to run Hermes at work. It can read your files, automate your browser, reply to emails, schedule jobs, and remember context across sessions. It is, in other words, exactly the kind of tool your IT department wants to say no to. They are not being unreasonable. An agent with terminal access, browser sessions, API keys, and long-term memory is not a chatbot — it is an employee-shaped problem for a security team. Hermes can do real work. That means it can also do real damage. This post is t”

“I pretty much only use Pro and I've yet to spend $20 in a month. It isn't my only model, but I use at least a billion tokens each month and it's far and away my most used one. It's my default for everything.”

“I asked chatgpt what it would cost for the same amount of tokens from other ai models and the pricing difference is insane. Ofcourse the value will vary with input and output tokens and even more so with the cache hits but even still ~11$ is insane price for a Billion tokens.”

“Crazy time wherethis kind of knowledge is available for free . Thank you David !”

“depends. If you use deepseek it literally cost peanuts. I had setup one of my agents to use deepseek v4 flash as default for mundane tasks and escalate to Claude Code or codex for complex coding etc however I realise that I had it deactivated and the whole time I was only using deepseek for everything. While it's obviously not as smart (I have not had it one shot any small, medium projects) but it does seem to get the job done if you're willing troubleshoot with it (by troubleshooting, you can l”

“Hermes Agent has been running on my machine for a week, 24/7. Every session, after it finishes a task, it reviews its own execution, writes what worked and what didn't into a markdown skill file, and uses that file the next time. You correct it once. It stops making that mistake. Six months in, it knows your preferred formatting, your stack, your tone in different contexts, and which shortcuts you want taken. How it fits against OpenClaw and Claude Code OpenClaw has gotten bloated. Updates consi”

“I am rechecking on hermes agent currently, also because many report great experiences, but oh my, does it look ugly. The web-UI uses such ugly fonts and background graphics, and for some reasons, UX feel slow and tedious (even in the tui). Pi mono agent feels quick and fast compared to it, and I can see immediately where it fails. Hermes seems to promise a lot more builtin features and to be more straightforward to solutions, but it feels sluggish in comparison. I use it with qwen3.6-35B and gem”

“I’ve tried openclaw, nanobot, nanoclaw, etc. but Hermes-Agent is my favorite! It takes all the features I actually care about with none of the bloat. More auditable/safe imo. Works well with my local models. I have hermes installed on one always-on mac mini with ssh access to my ubuntu servers via pre-shared keys. The mac mini is the “source of truth” or hub for all the orchestration. I use it either directly on mac mini, or ssh to mac mini from macbook, or via telegram. No multi-node setup! For”

“Installed this morning in less than 10 minutes. The installer migrated my OC agent and took care of the Telegram as well. I literally did nothing except configure the Ollama model. So far it's crazy fast and has home assistant support builtin. I fought with Open claw for days with memory problems and broken configs. I realize it's still early to post a good review but the setup was flawless.”

“Its amazing, its openclaw already set up and working, its like an OC with 1 week of debugging manually done + rag + memory persistence + better tool calling. (Qwen3.5-9b, 16gbvram), 10/10, only will go back to OC if it becomes at least on par with it”

“This software is extremely difficult to install for the average user, effectively locking many people out from using it. Unless you’re highly technical—ideally a developer or experienced coder—you’re likely to struggle or not be able to install it at all. The setup process feels unnecessarily complex and not user-friendly.”

“You can't properly uninstall this shit from windows. It's not listed in Software and components.Also I couldn't set lmstudio provider on setup screen - it couldn't "detect" it when I pointed the http://127.0.0.1:1234/v1 directly and wasn't allowing me to proceed without setting up any api key. Had to run hermes model in terminal for that and then it proceeded.”

“I’ve tried openclaw, nanobot, nanoclaw, etc. but Hermes-Agent is my favorite! It takes all the features I actually care about with none of the bloat. More auditable/safe imo. Works well with my local models. I have hermes installed on one always-on mac mini with ssh access to my ubuntu servers via pre-shared keys. The mac mini is the “source of truth” or hub for all the orchestration. I use it either directly on mac mini, or ssh to mac mini from macbook, or via telegram. No multi-node setup! For”