Today, I did something which got me feeling like the most advanced I have ever felt. I gave my AI assistant an AI assistant (imagine me with my hands doing the explosion sign on my head).
Okay I am overselling what I did. Let me give you some context.
I already have an AI assistant built on OpenClaw running on my VM server. It is pretty great. It is connected to Notion, Google Drive, Calendar, and a Gmail account I made purposefully for it. It can send emails, create calendar events, research topics, and run scheduled tasks like a morning briefing and so much more. I named it August.
But to be honest, sometimes it is like owning an old car that you are fond of but breaks all the time, so you are part time mechanic and part time driver. Not to talk about session bloating and API bills that creep up when you are not paying attention. Maybe I have to work more on my setup, or maybe I just have to accept that it is always going to be a fix-and-function assistant. Did I just coin a new phrase?
Anyway, then came Hermes. It kept showing up every now and then on my social media feed. The urge to test it was lukewarm at best, sitting somewhere on my growing list of things to try. And I am not kidding, that list keeps growing faster than I can work through it.
But one Twitter post stopped me in my tracks. I looked at it and thought, okay, I need to try Hermes for real. So I sat down, went through the process, and installed it on the same VM that OpenClaw is already running on.
Ladies and gentlemen, I am about to have the battle of the AI assistants. Who is your money on?
So What Is Hermes Agent?
Hermes Agent is an open-source AI agent built by Nous Research, the same lab behind the Hermes family of AI models. It was released in February 2026 and has been picking up serious momentum since.
Like OpenClaw, it runs on your own server, connects to messaging platforms like Telegram, and acts as a persistent personal assistant you can talk to from your phone. But the way it is built underneath is different in a few important ways.
The biggest one is memory. Hermes is designed around a learning loop. Every time it completes a complex task, it can write a skill document from that experience so it never has to figure out the same thing twice. It also maintains a structured memory file that gets injected at the start of every session, so it always knows who you are, what you are working on, and how you like things done. Over time, the idea is that it gets more useful the longer it runs.
It ships with over 40 built-in tools out of the box including web search, browser control, image generation, file operations, and a built-in cron scheduler for automated tasks. It also supports MCP servers, which means you can connect it to Notion, Google Drive, and other tools the same way OpenClaw does.
Why I Decided to Run Both Instead of Switching
This is the part I want to be honest about because the internet has a habit of framing everything as a replacement story. Hermes is not an OpenClaw killer, at least not yet and not for me personally.
OpenClaw has a much larger community. It has been around longer, has thousands of community-built skills, and my current setup with August has taken real time and effort to configure. You do not just throw that away because something new showed up on your Twitter feed.
What I decided to do instead was install Hermes alongside OpenClaw on the same VM and give it a specific job to do. I named it Lollie, gave it its own Telegram bot so I always know which assistant I am talking to, and pointed it at the same Notion and Google Drive that August already uses. No duplication, just a shared knowledge layer.
The migration story also helped. Hermes has a built-in command called hermes claw migrate that can pull your settings, memories, and skills from an existing OpenClaw setup. I ran it as a dry run first to see what it would bring across before committing to anything.
What Makes Hermes Different in Practice
A few things stood out during setup that are worth calling out for anyone considering this.
The install is genuinely simpler. One command, and it handles everything. OpenClaw takes more configuration to get running well, which is part of why the “part time mechanic” feeling exists.
The security defaults are better. OpenClaw has had a rough 2026 on the security front with several vulnerabilities disclosed publicly and thousands of exposed instances found running with weak default configurations. Hermes takes a more conservative approach out of the box with credential filtering and prompt injection scanning built in.
The cost model is more transparent. Hermes routes through OpenRouter which shows you exactly what each task costs per model. If you have ever been surprised by an API bill from running an always-on agent, this matters.
And the community web UIs are surprisingly polished for a project that is only a few months old. There is already a browser-based workspace called Hermes Workspace that gives you a full dashboard for managing your agent’s memory, skills, and tasks without touching the terminal. I am not using it yet but it is there when I want it.
Join The Wheels of Ai and Automation
A monthly drop of curated news, practical prompts, and how‑to guides—plus first dibs on workshops.
What I Am Still Figuring Out
I want to be straightforward here because this is a first look, not a verdict.
Lollie has been running for less than 24 hours as I write this. The memory and self-improving skills are things that need time to prove themselves. You cannot evaluate a system designed to compound over weeks in a single afternoon.
I also have not stress tested how the two agents perform on the same VM simultaneously. My server has 4GB of RAM and is already running n8n and a crypto paper trading bot alongside OpenClaw. Adding Hermes is another process competing for those resources, and I will be watching that closely.
The OpenClaw community is also significantly larger. With over 350,000 GitHub stars and thousands of community skills, it has an ecosystem advantage that Hermes, at around 22,000 stars, simply cannot match yet.
FAQ
Can Hermes and OpenClaw run on the same server? Yes. They are separate processes and do not interfere with each other as long as you give each one its own Telegram bot token and configure them on different ports. The main thing to watch is RAM, especially if your server is already running other services.
Do I need to choose between Hermes and OpenClaw? Not necessarily. They can serve different purposes. OpenClaw has a broader integration ecosystem. Hermes is stronger on memory and learning over time. Running both is a valid approach, especially during an evaluation period.
Is Hermes suitable for non-technical people? Not yet for setup. The install is simpler than OpenClaw but it still requires a server, terminal familiarity, and some configuration. Once it is running, the Telegram interface and browser-based workspace make day-to-day use accessible. But getting there requires some technical comfort.
The Takeaway
I do not have a winner to declare yet because this is day one. What I have is two AI assistants running on the same server, each with their own name, their own Telegram bot, and their own job to do.
August handles the workflows I have already built and trust. Lollie gets to prove herself.
Check back in a few weeks. The battle is just getting started.
Leave a Reply