My Personal Hermes Agent on an IONOS Virtual Server

I have been experimenting with AI assistants for a while. Not only as chat interfaces, but as working systems that can help me with research, writing, website changes, LinkedIn content, and daily professional routines.

At some point the question changed.

It was no longer: which provider app has the best chat experience?

It became: what would an AI assistant look like if it had my tools, my files, my context, and enough continuity to support me over time?

That is why I now run my own Hermes Agent on an IONOS VPS.

It is my personal AI assistant for my LinkedIn and blog that can prepare, research, draft, analyse, remind, and help me move faster without giving up control. Not as a fully autonomous content machine. Not as something that posts on my behalf while I look away. I stay in the driver seat.

Why not just use provider apps?

I still use provider apps. ChatGPT, Claude, Gemini and others are useful products. They are fast, polished, and improving constantly. But they are still mostly separate places. Each app has its own model, its own memory, its own tools, its own file handling, and its own way of interacting with the outside world. If I switch from one model to another, I often also switch the working environment around it.

With my own Hermes Agent, the model becomes only one part of the system.

The important part is the operating layer around it:

the same tools,
the same context,
the same local files,
the same recurring tasks,
the same channels to reach me,
the same workflows across different models.

That changes the experience.

If one model is better for writing, another for coding, another for long-context reasoning, I can choose more freely. The assistant does not have to lose its file access, workflows, or memory every time I change the model behind it. The value is not only model quality. The value is continuity.

The VPS is my agent’s home

Running Hermes on an IONOS VPS gives the assistant a stable place to live.

It is not tied to a browser tab. It can run scheduled jobs. It can work with files. It can interact with GitHub, my website project, my LinkedIn archive, and my own workspace. It can be reached through Telegram and email, which makes it feel closer to a real assistant than a single chat window.

For me, Telegram is the most natural interface. I can send a quick instruction from my phone:

Check my LinkedIn profile and suggest quick wins.

Or:

Draft three blog post ideas for next week.

Or:

Update the website copy on this section and create a PR.

Hermes can then use the tools available to it, inspect files, draft changes, run checks, and report back.

Email matters too. Some workflows fit better as mail: longer digests, scheduled briefings, or outputs that I want to store and process later.

The interface becomes less important. The agent is reachable where I already work.

What my Hermes agent supports today

The most useful workflows are not dramatic. They are the repeated ones. The small loops that would otherwise cost attention every week.

LinkedIn growth support

I use LinkedIn as part of my professional positioning around Enterprise AI, AI transformation, engineering leadership, and practical AI operating models.

Hermes helps me analyse my own presence. It can look at my profile, posts, articles, and positioning, then suggest quick wins:

where the profile could be clearer,
which themes appear consistently,
which posts support my positioning,
where the message becomes too broad,
which topics could be turned into stronger follow-up posts.

This is not about outsourcing my voice. It is more like having a sparring partner that keeps track of the bigger picture. I still decide what I want to say. I still edit. I still post myself. But I get a better starting point.

Daily AI digest with post drafts

I also run a daily AI digest workflow.

Hermes monitors current AI developments and connects them to my existing topics: Enterprise AI, operating models, evaluation, agent workflows, product ownership, and practical adoption.

The digest is not just a list of links. It tries to answer:

What happened?
Why does it matter?
Is there a connection to my existing writing?
Could this become a useful LinkedIn post?

When there is a strong angle, Hermes can prepare a LinkedIn draft and an image prompt that matches the design direction of my website and blog assets.

Again, the key word is prepare. I do not want an agent to auto-post industry commentary under my name. That would miss the point. LinkedIn is not a broadcast channel for generated content. It is a place to exchange ideas with people. Hermes helps me get from signal to draft faster, but the final judgement stays with me.

Weekly blog post ideas

Every week, Hermes suggests three blog post ideas.

This is more useful than it sounds, because good topic selection is often the bottleneck. I do not need random ideas. I need ideas that fit the arc of my work.

For example:

practical lessons from AI transformation,
why AI initiatives get stuck after the first pilot,
how companies should think about quality gates and evaluation,
why internal knowledge work is often a better starting point than flashy automation,
how engineering leaders can turn AI experiments into maintained products.

The agent can look at what I have already written and suggest the next logical topics instead of repeating the same angle.

That helps me build a coherent body of work over time.

Managing my website through the agent

My website is also part of the workflow.

When I want to change something, I can tell Hermes what I want. It can inspect the Astro project, update content, adjust copy, create a branch, open a pull request, and let GitHub deploy the preview version.

This is a very different interaction model from “open the repo, find the file, edit the component, check the route, commit, push, open PR.”

I can describe the desired change in plain language - from wherever I am. Hermes turns that into concrete work.

For example:

Make the homepage intro more focused on Enterprise AI operating models.

Or:

Add this new blog post draft and prepare the metadata.

The important part is that the workflow still has review points. A pull request is created. The preview deployment can be checked. I decide whether it goes live.

That is the pattern I want: faster execution, not less ownership.

… and many more

If I have a new idea, I can just ask Hermes to do it. It installs new skills, learns new things and remembers my preferences. So the next automation is just a quick message away.

The point is not maximum automation

Technically, much more automation would be possible. The agent could publish posts automatically. It could schedule LinkedIn content. It could update the website without a pull request. It could turn every trend into a post draft and push everything into a queue.

But I do not think that is the right goal.

For my work, the goal is not to remove the human from the loop. The goal is to remove friction from the parts that are repetitive, slow, or easy to postpone. I want help with preparation, analysis, drafting, research, and implementation.

I do not want to delegate judgement, taste, relationships, or professional responsibility.

Especially on LinkedIn, the exchange with people is the point. If someone comments, challenges an idea, adds a practical example, or shares their own experience, that is not an interaction to automate away. That is the valuable part.

AI can help me show up more consistently. It should not replace the conversation.

What this says about agents in general

Running my own Hermes Agent has also shaped how I think about AI adoption in companies.

Many organizations are still focused on individual tools. One team uses one assistant. Another team uses another. Some people upload documents into one provider app. Others build small scripts or internal prototypes. That can be useful, but it often stays fragmented.

The more useful question is what happens when an agent has a stable operating environment:

access to the right tools,
access to the right files,
recurring workflows,
clear permissions,
review points,
memory of relevant context,
and a human owner who decides what should happen.

That is where agents become less like demos and more like working systems.

My personal setup is small, of course. It is one person, one VPS, one assistant, and a set of workflows around my own work. But the pattern scales conceptually.

The hard part is not making the agent do more. The hard part is deciding where it should help, where it should stop, and where a human decision is required.