By Rolf Versluis · Published [DATE] on Uncommon Knowledge in Business and Life

A Hermes agent is a personal assistant. It runs tasks for you while you're driving, while you're asleep, while you're in a meeting. It is different from ChatGPT or Grok in one way that matters: it knows who you are, what your businesses are, what it is supposed to do for you, and it keeps doing those things, day after day, until you tell it to stop.

I have four of them running right now. One in Alpharetta, one at the farm in Chatsworth, one on a cloud server I rent by the month, and one on a Mac Studio in my office. They all talk to me through Telegram. They each have a profile of who I am and what they are for. They run while I sleep, they pick up work where I left off, and they do not need me to keep asking.

That is the part I want to explain. Because the word "agent" is doing a lot of work right now, and most of what it gets used for is not what I mean.

What I mean, and what I don't

A Hermes agent is a piece of software you install on a computer. It runs a large language model, the same kind of model that powers ChatGPT, but instead of waiting for you to type a question into a chat box, it has a job. The job is in its profile. The model reads the job, decides what to do, uses tools to do it, and comes back with the result.

Public cloud AI like ChatGPT or Grok is capable. I use it. But it only works when you are actively typing into it. When you close the tab, it is done. Yesterday's work is gone. So is the context about your business.

The agent runs in the background. You can text it from your phone the same way you would text a person. It reads what you sent, decides what to do, and either does it or asks you a question. It works through Telegram, Slack, Microsoft Teams, or SMS, whichever one you already use. It runs 24 hours a day. It can do research on the internet, write documents, create project plans, build or customize software, and do this without you watching.

The other part that matters: it can run locally. A Mac Studio with enough memory can run the model itself. Your data does not go to a cloud server. For some businesses (health care, financial services, anything with sensitive data) that is the whole reason to bother.

What it knows about me

Every Hermes agent has a profile. Mine says I am an engineer in Alpharetta. It lists my businesses. It says I am working on a chicken farm in Chatsworth and a small AI consulting practice. It says I am working on a wooden sword audiobook and a YouTube series and about six other projects I have not finished yet.

It also has a profile of what it is supposed to do. Some of mine do research. One of them helps me write code. One of them runs the surveillance system on the farm, which I will get to.

When I want to do a new project, I do not open a chat window. I text my agent. Sometimes I record a voice memo while I am driving and send it over. Sometimes I type a few sentences in Telegram. The agent reads what I sent, decides what to do, and either starts working or asks a clarifying question.

That is it. That is the entire workflow.

An example

I bought an IP security camera for the farm and mounted it on the building. I texted my agent about it. I told it what I wanted: when somebody drives onto the property, I want a notification with a picture. I want to know if it is someone I know or someone I do not.

We went back and forth for a few days. The agent did what it always does with new problems: it looked for an open-source project first. There is a leading one called Frigate. It would have worked, mostly. But it did not do what I wanted.

So the agent built a custom one. Took a few months. Now when somebody drives onto my property, the system figures out who it is, takes a picture, and sends it to me on Telegram. There are no false positives. A branch blowing in the wind does not trigger it. A shadow does not trigger it.

I have used cell-phone-connected trail cameras before. They send you a picture every time anything moves. They are useless for the same reason. Too much noise, not enough signal.

There is no way I could hire a person to watch my security cameras 24 hours a day. But I have an agent that wrote a program to do that, and the program runs without pay, does not sleep, and does not quit.

How it actually works over time

A new project starts with a conversation. I describe what I want. The agent writes a project plan. We break the plan into steps. We work through them, one at a time. Sometimes the plan is well-defined because I have done the work before. Sometimes it is not, and we figure it out as we go.

The other thing that happens, and this is the part I think most people miss: the agent builds up skills. A skill is a note it writes to itself about how to do something. The first time I had the agent install nginx, I had to walk it through every step. Now it has a skill for installing nginx, and the next time I ask, it is 99% of the way there before it starts.

One thing I want to be specific about, because it changed how I work. When I do something that has to be repeated, I do not have the agent do it. I have the agent write a computer script to do it. If you create a script, it is done the same way every time. The agent is fast. The script is consistent. Different things matter.

This is the part that accumulates. After eight months of working with an agent, you have hundreds of these skills. The work gets faster. The agent stops asking the same questions. You stop explaining the same context. The whole system gets more useful the longer you use it.

What it doesn't do

The agent does not make decisions for you, spend money without asking, send messages to customers without showing you the draft first, or replace the people who work for you. It does not take the calls. It does not have the relationships.

What it does is take the loops. The report you keep meaning to build. The follow-up email you keep forgetting. The vendor that needs chasing. The piece of open-source software that needs configuring. The camera that needs classifying. The website that needs updating.

That is the work I want off my plate.

What this is good for

When I run into a problem, I text the agent: "this is probably a common thing, how do other people solve this?" It goes out and does research, and comes back with a few suggestions. Sometimes it is a configuration fix. Sometimes it is a tool I did not know about. Sometimes the suggestion is "you are going about this wrong, do it this way instead." The agent does not have to know the answer already. It knows how to find the answer.

The way I think about it: this kind of working can make you five to ten times as productive, and it lets you do things you could not do before. If you think in organized ways but the detail work is not your favorite part of the day, you can have the agent do the detail work or set up a script that does it. That is the point. You bring the judgment. The agent brings the speed and the willingness to sit through the boring parts.

Why I'm writing this

I run a small consulting practice with my son Grant, who is a software developer. We implement Hermes agents for owner-operators of small and medium businesses. We pick the work, build it, hand it off, train the team. North Georgia and the I-75 corridor. Alpharetta, Chatsworth, Chattanooga, anywhere in between.

If you run a business and you have a few loops that eat your week, the first conversation about whether an agent fits is free. Send me a note at localhermesagent.com. Thirty minutes about what you are doing and what an agent could take off your plate. If it makes sense, we will figure out the next step. If it does not, we will tell you that too.