From OpenClaw to Hermes Agent: Why the Switch Was the Easy Part

The reason adding Hermes Agent to my setup went so smoothly had almost nothing to do with Hermes.

I keep a little AI agent running that I talk to from my phone like a coworker. It researches, drafts, helps with social media for my products and pokes at my code.

The newest version runs on Hermes Agent (by Nous Research) and getting it going was refreshingly boring. But the smooth part wasn’t really about Hermes — it was about groundwork I’d laid months earlier. Here’s the arc.

The OpenClaw experiment

A while back, when OpenClaw was the agent everyone was talking about — it went through a wild triple rename in about a week, from Clawdbot to Moltbot to OpenClaw — I gave it a shot. I spun it up on a VPS and used it for maybe two months to help manage and research social media for my products.

Honestly? It wasn’t great. But it did keep the ball rolling — drafts got written, ideas got researched — and I learned a lot about how I actually wanted to work with an always-on assistant. After a while, I switched it off, downloaded it’s personality etc in a .zip folder hoping to set it up when it got decent. Experiment over, but not wasted.

Building a home base

Then I revived an old HP laptop and instead of renting another server, I turned that into my always-on box — a little “home VPS” humming behind my TV. I wired it up so I could reach it from my phone over Telegram, with Claude Code as the engine. (I wrote about that setup in an earlier post)

That became the real foundation. I used it daily and kept expanding what it could do — and, crucially, it gave me a repeatable pattern for adding new tools to a machine I already trusted.

Adding Hermes

Recently my feed filled up with people raving about Hermes Agent — how capable it is, how well it remembers things, how nicely it runs unattended. So I decided to add it to the same home box.

And this is the part I want to highlight: it just slotted in. The always-on laptop was already there. Telegram was already my way in. I already knew how I liked to work with an agent. I’d even kept my old notes and memory files from the OpenClaw days, so I dropped those straight in. None of it was a fight — I was plugging a better engine into a frame I’d already built.

I’ve been using DeepSeek V4 Flash as the brains of the agent & also as the coding agent and I have to say, I’m genuinely impressed. For a model that cheap, it does really well. I built out a whole product concept and barely cracked 3% daily usage(more on this soon). I highly recommend it for personal tasks. Of course, be careful with secrets, environment files etc as you should with any AI Agent.

What I use it for

Three things, mostly:

Research while I write — I bounce ideas and outlines off it from my phone, mid-thought.
Social media — it drafts posts and flags things worth replying to, but never posts publicly without my OK. It drafts; I approve. It also drafts blog posts and recommends SEO updates based on how we are performing on G4A & the search console.
A few coding tasks — small fixes and scaffolding for the things I’m building; when something’s heavy, it hands the work to a dedicated coding agent and reports back. I mostly use the Claude Code setup for this.

The real win

If there’s a lesson here, it’s this: the smooth switch wasn’t because Hermes is magic (though it’s genuinely good). It was because past-me had already done the boring groundwork — an always-on machine, a way to reach it from anywhere, and a clear sense of how I work with these tools.

Build a simple, solid home base, and every new tool just clicks into it. New engine, same frame, zero drama. Smooth sailing ⚡