Autonomous AI agents like Hermes represent a paradigm shift from passive chatbots to proactive systems that continuously learn and improve through self-improvement mechanisms. These agents operate on core pillars including memory (personalized context), skills (predefined task playbooks), and scheduled automation (cron jobs), enabling them to execute complex workflows independently. The key differentiator is their ability to codify learned tasks into reusable skills, making them progressively more effective at specific user workflows over time. This self-improving capability transforms AI from a reactive tool into an autonomous assistant that can manage tasks like daily briefings, project tracking, and information aggregation without requiring constant human intervention.
深掘り
前提条件
- データがありません。
次のステップ
- データがありません。
深掘り
This System Saves Me 10 Hours a Week追加:
[music] >> If you are a working professional, irrespective of the job you do, every Monday morning you end up spending 45 to 1 hour of your day getting back to the speed.
You have lot of unread Slack messages, emails, Jira tickets, Confluence, and the list just goes on.
Someone updated the road map. Someone else raised a blocker.
Now, you have n number of Slack threads which you need to read before you even start your stand-up.
Now, none of the AI tools that we have currently help with this because they woke up this morning knowing absolutely absolutely nothing about you. You have to brief them before they can brief you.
But, what if there was other way around?
What if your AI already knew your projects, your team, your priorities, and every morning before you even open your laptop, it has already pulled out everything together and sent it straight to your phone on Telegram or WhatsApp?
That's not a fantasy. That's Hermes, and that's exactly what we are going to set up today.
Uh we'll start with a little bit of uh theory um and I'll spend like next uh 5 to 10 minutes in terms of explaining what Hermes is and uh how does it operate.
And then we'll get into the actual uh set demos or setting. So, Hermes is an open source AI agent built by Nous Research. It's an MIT licensed, which means it's completely free to use and modify.
And as of right now, it has over 140k stars on GitHub. So, that's not a small project. That's one of the most popular AI agent repos in existence.
But, raw numbers doesn't tell you the whole story. Here's what makes Hermes actually different.
Now, couple of things to know is first of all, it can run on any infrastructure that you have. So, it can run on your phone, your laptop, uh it's a virtual VPS, uh Docker, or even if you have an Android phone, you can run it uh via something called Termux. You are not renting someone else's cloud brain.
You own the whole thing.
Now, moving further, Hermes can be used anywhere.
So, basically, it can it it can talk to your 15 messaging platforms.
Uh you can set it up on Telegram, Discord, uh Slack, WhatsApp, Signal, uh iMessages, emails, SMS. Basically, wherever you already live, you pick the app, Hermes shows up there.
And interesting thing to know about Hermes is uh And and this is something which sets us apart. So, it's self-improving. Uh it improves over time by writing its own skills. Every time it completes a complex task, it codifies what it has learned into a re- into a reusable file.
So, it gets faster and smarter at helping you specifically the more you use it.
But, does it not look like like Claude Code or Open Claw? How is it different?
So, it's a really good question to know. Uh they're all agents, but one thing to know is they are pretty different. Uh So, when you look at Claude Code, it's an Anthropic coding assistant. It lives in your terminal right next to your code. You are actively driving it. It's best for when you are sitting at your keyboard doing development work. Or in worst cases, like if you are away, uh you still need to keep your monitor awake uh to continue using Claude Code. Open Claw, I think, slightly different. It's a fully autonomous agent. Uh, more than 300 3.5 lakh GitHub stars, huge teams, tons of features.
And Nvidia actually built an enterprise version on top of it called Nemo Claw.
Very capable, very broad.
But Hermes is different in focus. It's bigger, faster, and specifically built around one idea, self-improvement. It was built for thinkers.
It loves open models like Gwen and Lama.
It literally rewrites its own skill as it learns, updates, shifts, and modifies it to time.
In short, while Cloud Code helps you code while you are working, Open Claw is powerful multi-purpose agent. Hermes the one that gets better at being yours the more you use it.
Now, I think here's [snorts] the meaty part, and I'll I'll promise not to make it very complex. So, if you look at Hermes, it's built on five core pillars, and it's very important to understand when you operate with Hermes. So, first of this is the memory, and uh, so it's small and durable. It's not a transcript of every transactions or conversations you have ever had. It's the things Hermes generally need to carry with it every time it talks to you. So, it typically lives in two files, user.md. It's more about you, who you are who you are, your style, your preferences.
Uh, basically, you can think of it as your personal profile that Hermes reads at the start of every session.
Memory.md is about your environment, your projects, what your setup looks like, and the facts that Hermes has learned about how you work.
And both these both of these get loaded at the start of every conversation.
And they take snapshots time to time.
Hermes reads them, gets oriented, and then helps you.
Next part is skill.
Um so, you can consider skill as So, so the difference between memory and skill is let's take for example if you have to prepare a dish.
Memory is something which you know that hey, if you have to prepare a dish, uh you need to do so and so things that you have learned from the memory.
But, skill is primarily predefined task that hey, if you have to prepare let's take for example dosa, right? Uh it it should be done in x number of steps uh so that you actually create a really good dosa. Consider this as a self-improvement engine and it's generally one of the coolest thing about Hermes.
Uh skill live at a skill.md md file and it contains instruction for how to do a specific type of task.
Uh consider this as a playbook.
Now, moving forward there is soul.
This one is optional and but it's kind of useful. So, you will you will see a file called soul.md and it's literally the personality file of your Hermes.
Uh then comes crowns. Uh these are primarily like schedule automation. You will You would have seen this in cloud.
Uh Hermes also has this capability so they turn Hermes from a tool which you talk time to time into an into an autonomous agent that does the things automatically on a schedule uh wh- while you are kind of away doing something else.
So, let's take for example if you want to you if you want to set up certain automations to cert- certain automations to brief you about the latest AI news every 6:00 uh early morning when you start your day. So, you You set this up in a natural language, uh on Telegram. So, every morning you wake up, this cron will run and it will send you updated news about news about uh latest AI news.
So, a few things to know about how cron works under the hood because this matters. So, each cron job runs in a fresh isolated session. That means it doesn't inherit whatever you were talking about in your last chat. So, it starts fresh, reads your memory file, and does the job. The output gets delivered to you be it in Telegram or a local file or any other connected platforms.
And this is what ties all the four pillars together, and we call it loop. Uh I think it's worth spending a couple of minutes here. So, Hermes So, in terms of if you look at the whole cycle, Hermes does the work with its tools. It saves facts to your memory files. It turns repeatable steps into a skill for the next time. It searches the past session when old context is relevant. And then it goes to back to step one. So, basically, this is the way it helps or it completes the end-to-end loop.
Every lap of this loop makes Hermes more effective at helping you specifically, not just better in personal, better for your workflows, your preferences, and your projects.
To get started, just search for Hermes agent in Google search. And uh you would see it in the search result. So, yeah, just click on here. I see that interface has changed recently.
Now, you can go through like if you want to read more about it, go to the docs.
Uh there's portal, there's desktop apps, and there's there are social media channels as well.
Now, with regards to installation, uh you can install it on your desktop. Uh it supports both Mac Mac and Windows.
And then you can also set up uh virtual OS via Hostinger. So, in today's today's discussion, uh I'm going to install it locally on my Mac, and then we'll try to connect it through a Telegram and set up an automation workflows.
So, let me just quickly copy this and run this. In my case, okay, sorry. In my case, I have already installed, so sorry.
Uh Let me open the terminal.
Okay.
Uh So, yeah, it had just started processing.
Uh you It will ask like which AI model to connect to, and there are tons of options.
I think in I I have not So, easiest one is OpenRouter because it connects to it can it's kind of an orchestrator where where it connects to multiple AI models.
In my case, I'm going to start with Anthropic Cloud because I have already um paid for that. So, let me select this.
And uh yeah, it's going to ask like whether I want to use existing credentials.
Since previously I've installed this, so the credentials are stored.
Uh I don't want to reauthenticate, so I'll just go ahead with option one and use existing credentials.
Uh Now, in terms of model, again, this this is where you have to be careful, like if you end up selecting Opus 4.0 or 4.7, it's going to consume your token really fast. So, I'm going to select Sonnet uh 4.5 And in terms of space uh So, there are multiple options whether I want to run it locally, Docker model SSH or Daytona. So, I'll go with my current setup which is current local.
Now, I think at this point of time it is actually asking me if you want if you want to connect with any messaging platform and all and let's try to do it with Telegram.
uh So, in this case Telegram is already configured but I'll try to reconfigure it. And in this case I'll select as no.
uh Okay, so Telegram already configured reconfigure.
So, let's start with reconfiguring Telegram.
Now, when it comes to Telegram the way it works is you have to set up a bot and then uh there's a bot token. So, basically you have to uh you have to actually share the bot token here. That's how Hermes will connect with the Telegram.
So, let me log in to the Telegram first.
uh The way things work in Telegram is uh there is already a bot which which is called Bot Father. uh Using that Bot Father you can create a new bot. So, as you can see here I mean you can search here uh Bot Father. Once you search it it will actually direct you to uh different options. So, let's take for example if we want to create a new bot I'll just type / uh new bot and it will guide me through creating a new bot.
um Okay, let's call it Hermes test.
Good, let's choose a username for your bot. It must end in a bot. Okay?
Uh Username could be Yeah, I mean, let's go with one example.
Register bot.
Okay, so this username is already taken.
Let's think of something else. Uh Genie.
Oh.
Great. So, this is the bot name because we need to know and keep a record of all of this because Telegram is going to ask this.
So, as you can see in this message, congratulations on your new bot. You will find it at uh this. You can add a description about section and profile picture for your bot, blah blah blah. And this is the token for accessing Hermes bot.
Now, let's click on this one.
Okay?
Now, we'll go back. So, it asked for create a bot via BotFather on Telegram.
Telegram So, as you have seen, we had created a bot and there was a bot token which we have got. So, I'll just copy and go back.
And Okay, 1 second. Let me paste the bot token.
Uh okay, so it in terms of security, it is asking to find your Telegram restrict who can use your bot to find your Telegram user ID, message user info bot on Telegram.
And it will reply with your numeric ID.
These informations are needed because you you just want to make sure that 1 second, just hold on.
Hermes needs user ID I or user info to connect with your Telegram.
So, just message user info to user info what and it will come up come back with user IDs.
So, let's a lot user IDs comma separated. Okay, so let's input the user ID.
Uh Okay, now I think couple of questions that he has asked is cross platform messaging notifications for Telegram DM.
This is your user ID. Same as above. As your as your user ID as the home channel.
Uh okay, I think one thing to figure out is do we want a separate user ID for uh this particular channel or not?
I think it should be fine. Uh we can go ahead with selecting the main user ID.
And uh I think yeah, messaging platform is configured. Restart your gateway to pick up changes.
Uh okay, so let's restart.
Uh it is asking me configure CLI, configure Telegram, uh configure all platforms. I think I want to skip all of this and mark this as done.
Uh once the setup has been done, you can click on help me chat uh and then it will start this chat window. You can chat here as well. Let me ask a few questions.
Uh okay, what is 2 + 2?
And this will be just a basic test to understand if it Hermes is uh working or not. So, it has come up with an answer of four, okay. What you can help me with?
Okay, so it's it's coming up with whole lot of answers in terms of I'm Hermes, a CLI agent that can help you with blah blah, okay.
Uh Can you check Can you check if my Telegram is set up?
Okay, so yes, your Telegram is set up and connected to Hermes. I can see one Telegram target available.
Uh Okay, so we can ping Telegram from here itself. Say, "Hello to Telegram bot."
Uh I want Hermes Hermes Jenny bot. So, I'll just see if it is able to ping the bot that we have created just now or not.
Okay, so now we'll go back and check in Telegram if uh it has received a hello from Hermes or not. So, I don't have a hello from Jenny Hello from Hermes.
Now, I would like to set up an automation. I think uh I think one thing which I can think of is First of all, if you are setting up any automation, don't give a very open-ended task. Try to give them a more defined and guided task because what happens is sometimes like if you get give them a open-ended task, it goes into processing and then it can fail. Right? So, I will just start with I want to set up an automation automation for uh keeping myself updated with latest AI latest AI news. Now, when it comes to AI news, like there are multiple blog post, news.
Uh so, if we give this kind of open-ended information, it will go run through, and then it will end up end up your end up consuming your tokens.
So, I think updated with latest AI news, I want to run it 8:00 a.m. every day in the morning IST every day in the morning.
Uh and let's take for example, just to keep it simple simple, I want the news to be pulled from Google News.
Please set this up.
And just see what I miss comes up with.
So, as you can see, uh the task that we have given, it has started breaking this down and it has started defining in terms of what action it needs to do. So, uh it says that, "Hey, I'll set up a daily AI news briefing for you at 8:00 a.m. IST. Let me create a cron job."
because in order for it to run on a regular basis, a cron has to be created.
So, it has scheduled a cron job 8:00 a.m. IST, and then there is also a job ID associated to it. And next run is scheduled for tomorrow 8:00.
Uh okay, you can manage this automation anytime. Uh we can just chat, "Hey, pause, modify, remove." Want me to adjust anything about briefing format or timing? I would say, "No."
Uh this looks fine.
Please proceed.
So, as you can see, it has set up uh and given us given me a confirmation that your uh cron job for scheduling scheduling AI news is set up.
I think what I want to eat want to do is I want to run a test. So, I'll ask, "Can you can you run it now as a test?"
So, yeah, it will start processing and in this case it has triggered.
Perfect. So, as you can see, the test run was triggered and it has pulled the key uh our top AI news from Google search and it has directly kind of delivered to my uh Telegram.
Uh so, this is just incredible, right?
Uh I mean anything that you want, it can be delivered to your uh Telegram. Or even you can set You should be able to set up WhatsApp as well.
Uh but the key fundamental is like Hermes is far better than any other agentic framework that you have been currently interacting with.
Uh and then it improves over time, right?
You just need to start trying it out. Uh set up couple of automations and see how how how it operates.
Now, before I kind of conclude, I I also want to show you some things.
Uh okay.
Just hold on.
So, if you click on Hermes uh docs, you can see here user stories and use cases.
So, these are like different use cases that different users have solved and built for. You can take inspirations uh from this. If something is actually relevant for you, you can just try to copy-paste or probably just redo the same thing and uh experience it firsthand.
So, that's pretty much everything about the basics of Hermes and setting up a automation in Telegram.
Do let me know your feedback, and if you want to kind of experiment something else, and share with you guys. I will be happy to know about it.
関連おすすめ
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30











