Home of a Hermes Agent Enthusiast

I’ve been using Hermes Agent since July 2026, spending more than 12 hours a day tweaking, tuning, and expanding it.

Feel free to ask me anything. Let’s make Hermes great!

And don’t forget to check out my blog! 😜

My World Changed on July 17, 2026. Here's Why.

That day I installed Hermes.

I've been using artificial intelligence since late 2022, when the first ChatGPT came out — but it's the Hermes agent that improved my way of communicating with AI a thousandfold.

Hermes Agent

The easiest way to understand what Hermes does is with an example. Say you want to convert a Word file to PDF. You ask Hermes how to do it.

Hermes breaks your task into parts: Word, converter, PDF. First it clarifies the details — what kind of document you have, what tables are in it, whether the PDF needs to look exactly the same. Then it starts figuring things out: how a Word file is structured, what conversion methods exist, which one is best. It tests options right on the server — installs a converter, tries it, looks at the result. And in the end it converts your file, shows you the result, and if something is wrong — it redoes it.

It turns out Hermes isn't just a program — it's a whole employee. It has three key qualities. First, a brain — a powerful neural network that can think and plan. Second, hands — it runs on a server where it can execute commands, read and create files, browse the internet, and talk to any API. And third, memory — it remembers your entire correspondence, all your projects, and all your preferences. It doesn't forget them an hour later like a regular chatbot does. The brain thinks, the hands do, the memory never fails.

Imagine this: someone sent you an add-on for 1C:Enterprise, but it doesn't fit your configuration. Hermes can take that file, go to the developer's website, read the documentation, dig into your 1C code (via configuration files or API), figure out what the problem is, and then rewrite the add-on's program code so it fits your configuration.

You can connect anything to Hermes — other neural networks, image generators, search engines, email, CRM. With every new tool it gets smarter, and its capabilities grow wider.

As a result, Hermes becomes an orchestra conductor: it reads your email, checks tasks in your CRM, prepares replies, verifies information on websites, finds the best prices, creates price lists or ad feeds. From a technical standpoint, no task is impossible for it. And since it runs on a server, it works around the clock — even while you sleep. The perfect assistant!

Rule Zero

If you've already realized you need Hermes, the most important thing is this: treat it as another person — very professional, but still unable to read your mind. Just as you would brief another person before assigning a task, you need to tell Hermes everything.

It's silly to ask it "create an advertising report for Q1 2026." Instead, you should write: "We sell cucumbers. We have our own greenhouses, we grow 40 tons a month. A vegetable wholesale base buys 10 tons. The other 30 tons we need to sell aggressively, because there's heavy competition from gardeners. Our cucumbers cost 40 kopecks, theirs — 30. We run Google contextual ads — I'll send you the export of our account. Look at the conversions and reach. We need to seriously boost sales, 4-6 times, because in a week the cucumbers will rot and there'll be nothing to sell. Right now I'm interested in analyzing the bottlenecks of our advertising. If you need more numbers, say so, I'll send them."

That kind of prompt is a great way to start a conversation. And to actually solve the problem, you'll need no fewer than 20 such dialogues. Get ready for a long communication that you'll keep coming back to over time.

Rule One

Install Hermes on a server. Yes, you need to buy hosting (a dedicated server) and put it there. If you install it on your laptop, you lose 50% of its capabilities in that same second. Hermes must work always, even at night. A server is its natural home.

Rule Two

Connect as many resources to Hermes as possible. Connect DeepSeek, Claude, Qwen. Connect built-in image generators and external search engines like Brave and Parallel. Yes, it costs money, but it's pennies — they pay for themselves the same day.

Rule Three

Connect it to your Telegram. Simple questions and tasks you can comfortably discuss and assign there. It saves a lot of time. Remember: Hermes stores your entire correspondence in a database and has full-text search — it will instantly find anything about saccharin that you discussed months ago across 100 dialogues.

Rule Four

Create a project folder for each task. If you just need to check currency rates, ask in the Telegram chat. But if you need to process documents, discuss a long-term task, and then make daily edits to it — you need a folder on the server and working with Hermes in the console (CLI). Having separate folders for each project prevents chaos: it's convenient for both you and Hermes, nothing extra in sight. Hermes doesn't confuse projects and doesn't "pull things by the ears."

Rule Five

Train it. When it does a task well, write: "Well done, great job." This reinforces its learning. Step by step, it starts to understand what result matters to you. Next time it will be smarter. Also don't forget to remind it: "We'll do this task every Friday now." This helps it keep its toolkit instead of reinventing it from scratch every Friday.

Rule Six

The Hermes-Agent project is actively developing — update it constantly, at least once a week (updates come daily). This reduces bugs and keeps growing the functionality.

In Conclusion

Hermes genuinely can do anything. Be patient and tell it as much as possible about your task and the result you need. It will work magic.

If you need detailed consultation, implementation, and setup — write to https://t.me/AndrewGolubev