← writing

From tools to systems: my journey with Hermes Agent

I didn’t start with Hermes. I started where most indie builders do - jumping between models, comparing costs, falling into “best cheap LLM” rabbit holes, trying to stitch workflows together manually.

It worked. Barely. Every setup felt temporary. Every workflow broke after a few days. I had tools, but no operating layer.

The question that changed things

When I found Hermes Agent, the question shifted. Not “which model is best?” but “how do I build a system that can use any model?”

That’s a different problem to solve. And it’s a better one.

Hermes isn’t a chatbot. It’s a persistent AI agent that runs on your own infrastructure - VPS or cloud - with memory across sessions, tools, and the ability to automate and delegate. It behaves closer to a real assistant than a prompt interface.

Three things that made it stick

First: it lives outside any chat window. Hermes runs via CLI, Telegram, background processes, cron jobs. It works even when I’m not actively in a conversation. That’s a meaningful upgrade from “open tab, ask question.”

Second: it’s model-agnostic. I can switch between OpenRouter, OpenAI, Anthropic, or Nous Portal without rebuilding anything. So I use cheaper models for routine tasks and upgrade only when the work actually needs it. That alone fixed most of my cost anxiety.

Third: it does real automation, not just text generation. Terminal commands, file read/write, web search, scheduled tasks, sub-agents for parallel work. That’s when it stopped feeling like an assistant and started feeling like an operator.

What I actually use it for

Day-to-day, Hermes handles ads reporting workflows, email and newsletter summaries, Telegram-based task interactions, and chunks of my builder workflow - coding help, small automations, acting as a second brain when I’m stuck.

It’s not perfect. But it’s reliable enough to depend on, which is more than I could say for everything I tried before it.

What didn’t work

Setup friction is real. CLI config, API keys, config files - if you’re non-technical, you’ll struggle. I struggled. It took a few sessions of debugging before things ran cleanly.

It also doesn’t fix weak models. Hermes gives structure to whatever model you’re running. A bad model with good structure is still a bad model. And if you only think in prompts rather than workflows and state, it’ll feel confusing at first.

The real shift

Before Hermes, I used AI tools. Now I run an AI system. That changes the question from “can AI do this?” to “how do I design this workflow so AI can handle it?”

That’s a more interesting place to work from.

I’m still early. But the direction is clear - build repeatable workflows, reduce manual work, turn systems into products. Hermes isn’t just a tool in that picture. It’s the foundation layer.