Hermes: The AI Assistant That Actually Gets Things Done

When we talk about AI, most of us picture something like ChatGPT. Open a chat, ask a question, get an answer. If you need something else, go back and ask again. But what if the AI didn't just sit there waiting for your next question? Imagine an assistant that knows your schedule, keeps track of appointments, sends messages, follows up on loose ends, finds information across the systems you use, checks your calendar, and can even do certain things on its own under rules you've set. That's what Hermes has become for me. I've been using Hermes as an AI assistant for a while now. In this article I want to explain what it is, how it differs from a chatbot like ChatGPT, what it can do, how I actually use it, and why giving any system that kind of access calls for a lot of care.

What three months with Hermes have really felt like

I've been using Hermes for more than three months, and honestly, I'm really happy with it. It feels like having an all-purpose assistant that takes a lot of work off my plate and keeps it moving. That's not a feature-list claim. On busy days, it has genuinely made my life easier. It has limited access to some company systems and sites I use. It checks university pages, reminds me about unfinished work, and if I say, “Follow up with this person,” it can message them from its own account, get a reply, and tell me what happened. Sometimes it even asks how I'm doing and understands what has been taking up my time lately. I might say, “See if the university has announced anything,” or “They still haven't replied about that thing; follow up again,” or “I'm swamped today; remind me of the important stuff.” The part I love is that I don't have to explain the whole backstory every time. I can say, “Follow up on what we talked about yesterday,” and it usually knows what I mean. Sometimes I've forgotten a task and Hermes still remembers it. Before this, AI mostly meant asking a question and getting an answer. Now it feels more like someone beside me takes a task and actually moves it forward. Of course, when a task involves another person or a sensitive system, it still needs clear rules and limited access. Concept image made for this article, not a screenshot of Hermes.

So what is Hermes?

The simplest way I can describe it is this: To me, Hermes is like a digital secretary or personal assistant that can always be available. The key difference is that you don't just talk to it. You can give it access to tools, systems, and services, then ask it to use them to get work done. For example, your assistant might:

manage your calendar;
find important emails;
keep track of conversations;
message a specific person;
check information in a system;
watch for a particular change;
remind you at the right time;
or carry out a sequence of steps for you.

That's why I think of Hermes as an AI agent more than a chatbot. The Hermes Agent site describes its memory, tools, messaging integrations, and scheduled work.

Official Hermes Agent artwork. Source: Nous Research.

How is Hermes different from ChatGPT?

ChatGPT is powerful, and I use it too. But the way I use Hermes is different. With a regular chatbot, the interaction usually starts with you. You open an app or site, explain a problem, and get a response. An agent can connect to other tools and, when it has the right permission, use those tools. Say I tell it, “Let me know if anything important changes in my university schedule.” Normally, I'd have to keep checking the university site myself. With the right access and tools, an agent can check the system and bring a change into my workflow. Or I might say, “Follow up with that person about this issue.” If the assistant has access to a specific messaging account and I've set the rules for it, it can send the follow-up. That's the difference that matters to me: a chatbot talks with you; an agent can also use tools and do work.

A view of Hermes Desktop. Image source: Ollama's Hermes Desktop guide.

Why do I run Hermes on a server?

Where an assistant runs matters. You can run an agent on your own computer, but then it's only active while that computer is on and connected to the network. I prefer to run Hermes on a server. The reason is simple: the server is almost always on. If something happens at three in the morning, it can still carry out a task I've set for it. I also don't have to leave my personal computer running all the time or keep its network setup ready for every integration. On a server, the agent feels more like a permanent service than an app that's available whenever I happen to open my laptop.

The most useful part for me: skills

One of my favorite parts of the system is the idea of a skill. I think of a skill as a way to teach your assistant how to do a particular job for you. Some skills are made by other people. Others can be completely personal. Imagine I travel a lot and have particular tastes. I can explain the kinds of places I like, what I avoid, my budget, and the sort of experience I'm looking for. Later, I can say, “I'm in a new city. Suggest a few good places,” without repeating every preference. The assistant knows what good means to me. The same idea works for dozens of things: tasks, travel, reading, university, planning, following up with people, writing, technical work, project management, or any process you repeat in your life.

What if there's no ready-made skill?

This is even more interesting to me. Suppose you want to connect the assistant to a system that almost no outside developer knows, like a particular Iranian university portal. The odds of finding a ready-made skill for that local system are pretty low. You can instead explain the process to the agent and give it the tools it needs. Show it where to find classes, where exam dates appear, how course information is displayed, and which details matter to you. A skill doesn't always have to be something a programmer wrote in advance. Sometimes you can teach the process through conversation and testing. The more sensitive the system, though, the more important it is to test it and limit its access.

What do I actually do with Hermes?

This is the important part. I'm not interested only in theoretical features. What makes Hermes exciting to me is what I've trusted it to do in real life.

1. Connect my life to Google Calendar

My calendar is one of the most important parts of my setup. Meetings, important plans, and events I can't miss go there, and Hermes can use that information while helping me plan. That turns the calendar into more than a page full of events. The assistant can understand some of the context too. It knows when a meeting is routine and when one really matters to me. That little difference makes reminders smarter.

2. Check my university system

My university is online, and class times, dates, and other plans can change. I've taught my assistant to check the information I need in the university portal and bring it into my planning. Instead of opening the portal over and over to see whether anything changed, I can hand some of that checking to Hermes. It takes a surprising amount of mental noise away.

3. Give the assistant its own Telegram account

One detail I really like: I made a separate Telegram account for my assistant. I deliberately didn't give it my personal account. Why? When I'm talking to someone, I want them to know it's really me. If my assistant sends a message, that should be clear too. So I can tell Hermes, “Ask this person about that issue,” and it can message them from its own account, making clear that it's following up on my behalf. That feels much healthier than letting AI secretly speak through my personal account while the other person has no idea whether they're talking to a person or an agent.

4. Reminders that are more than alarms

A basic reminder says, “You have a meeting at four.” I expect more from an assistant. Suppose the meeting is very important and I know where it is. The assistant can consider the location and roughly how long it will take me to get there, then remind me to leave early enough. One meeting might need a single reminder. Another might matter enough that I want several. So a reminder becomes something that understands the context of my day, instead of just ringing at a time I typed in.

5. Talk instead of typing

Honestly, a lot of the time I can't even be bothered to type. I send a voice message. If an idea comes to mind or I remember something I need to do, I can talk about it right then. That makes using the assistant feel much more natural. I don't have to sit at my laptop and write a neat, formal prompt every time. I can explain what happened and what needs doing almost the way I'd talk to another person. That matters a lot to the feeling of having a real assistant.

HERMES / CHATMy real chat with Hermes

00:00 / 00:00

A real screen recording of my chat with Hermes. The conversation and audio are in Persian.

6. Help with writing, code, and servers

My work isn't limited to my personal schedule. Sometimes I need help writing text or a script. Sometimes it's code. Sometimes we need to think through a problem with a website or server. Hermes isn't a single-purpose assistant for me. What it can do depends a lot on the model behind it, the tools you give it, and the skills you define. The better those tools fit the job, the more complicated a task it can handle.

7. Keep up with my workload

I've also set up something like a daily check-in. When I have several projects going at once, the system can use the rules I've given it to follow some things more closely. It doesn't just see a list of tasks. It has some context for what I've been focused on lately, what's fallen behind, and what needs another follow-up. This matters to me because the real problem is having several open loops in my head at the same time. An assistant that carries some of that working memory makes the day feel lighter.

Hermes and OpenClaw: why did I keep Hermes?

I also spent some time using OpenClaw. In my experience, OpenClaw gave me more freedom and possibilities in some areas, and I could do some really interesting things with it. That extra flexibility had a cost for me, though. It needed more technical attention. Sometimes I had to investigate bugs and spend time debugging it. That felt a little at odds with the whole point of having an assistant. If I'm regularly spending hours repairing it, I lose some of the time it was meant to save. In the end I kept Hermes in my main setup. That doesn't mean Hermes is better for everyone. Someone who loves experimenting with and extending agent systems might prefer a more open, complex tool. My priority was stability in everyday use.

Personality: more than a language model

Another interesting thing about agents is that you can shape how they behave. You don't just decide which tools they can use. You can also tell them how to speak, how formal to be, when to ask questions, when to be more careful, and which actions need your approval. Over time, it can feel more like dealing with a particular assistant than with an anonymous model. That personality should never make us forget that this is still AI. It can be wrong, misunderstand you, or make a poor decision. Even my own assistant says in its introduction that important work should be checked and tested. I shouldn't trust an AI output just because it sounds confident.

Concept image made for this article.

The most important subject: security and guardrails

So far, this all sounds exciting. But there's a serious question underneath it: isn't it dangerous to give AI this much access? My short answer is yes, if you give an agent unrestricted access, the risk can be high. That's why guardrails, the limits and rules around access, are one of the most important parts of my setup. I decide what it may do on its own, what it must never do, and what it can do only after I explicitly approve it. For financial or sensitive actions, I want it to ask first. Sometimes I even use more than one approval step. The idea is to give the agent room to work within a clear boundary. Good guardrails don't make risk disappear, but they can reduce it and make the system easier to control.

The more powerful the agent, the stronger the guardrails need to be

If an agent can only generate text, its worst mistake is probably a wrong answer. But once it can send a message, change a file, take action on a website, book an appointment, or connect to a real service, a mistake isn't just a bad response. It can become a bad action. That's a huge difference. The more capabilities we give the assistant, the more seriously we need to take controls, logs, permissions, and approvals. To me, the future of AI agents depends on getting that balance right: giving them enough freedom to be useful while keeping people in control.

Should we hand everything to an agent?

No. At least, I don't. Seeing an assistant handle several kinds of work makes it tempting to give it more access right away. I'd rather move step by step: one simple task, then one limited connection, then a review of what happened, then maybe another skill. Only when I trust a process do I increase the access it gets. I think extra caution is essential with money, personal information, important accounts, and actions you can't undo.

Who is Hermes for?

An agent won't be equally useful to everyone. If your days are fairly simple, a calendar, reminders, and ChatGPT might cover most of what you need. But if you're juggling several projects, talking to lots of people, dealing with meetings, balancing university and work, or checking different systems all the time, an agent gets much more appealing. Especially if these thoughts keep circling in your head: “Don't forget to check that.” “I need to message them again.” “Has the university announced anything?” “I can't miss that meeting.” “I need to follow up on that result.” That's when having something outside your own head hold part of the context becomes valuable.

What is Hermes really worth to me?

Its best feature isn't that it gives clever answers. Plenty of models can do that now. For me, the real value is this: Hermes has taken over some of the work of remembering and paying attention. I don't have to keep every university update in my head. I don't have to remember every follow-up on my own. I don't have to constantly wonder whether I've forgotten an appointment or a task. I'm still the one making decisions. But I've handed some of the watching, recording, reminding, and following up to another system. When you're dealing with several things at once, that can be incredibly useful.

The future agents are building

I don't think the next stage of AI is just models that talk better. Language models are already very good at conversation. The bigger change comes when that intelligence connects to real tools. When AI can notice something, recognize a change, make a limited decision, use a tool, and check the result, we're moving from chatbot toward agent. The main question stops being only, “What does AI know?” and becomes “What can AI do for me?” Then we have to ask the next question immediately: “How much should I allow it to do?” Those two questions belong together.

Final thoughts

If I had to sum up my experience in one line, Hermes sits somewhere between AI, automation, and a personal assistant. I can talk to it, define skills, give it tools, connect it to systems, and hand over some repetitive work and everyday follow-ups. What makes that useful isn't just the power of the underlying model. It's the combination of the right model + the right tools + the right skills + useful memory and context + sensible guardrails. Put those pieces together and an agent becomes more than something you ask questions. It gradually becomes part of how you work and live. I still don't think we should give AI complete control over our lives. In fact, the more capable these systems get, the more their access needs to be controlled. But within the right boundaries, my experience has been that an assistant like this can take away a meaningful amount of repetitive work and mental clutter. Maybe that's the biggest shift agents bring: we're no longer just talking to AI; we're learning how to work with it.