How to Set Up and Master Your 24/7 Self-Improving Hermes AI Assistant

How to Set Up and Master Your 24/7 Self-Improving Hermes AI Assistant

Visit Hostinger VPS now

Stop wasting time copying and pasting prompts into a browser tab. Most AI tools forget who you are the second you close the window. Hermes changes that by living on its own server, working 24/7, and messaging you first when something matters. It is a self-improving agent that connects to Telegram, Discord, or Slack to handle your life while you sleep.

Hermes is different because it writes its own skills. When you give it a hard task, it saves the process. The next time you ask for something similar, it uses that saved recipe to work faster. It also has persistent memory, meaning it remembers your preferences and work style forever. You can schedule jobs using plain English and talk to it across 15 different platforms through one gateway.

To get this running, we will use a VPS, connect it to an AI model provider called OpenRouter, and link it to Telegram. We will then cover how to train its memory and set up autonomous jobs.

Deploying Hermes on a VPS via One-Click Template

You could run Hermes on your laptop, but it would stop working the moment your computer sleeps. A Virtual Private Server (VPS) is a cloud computer that stays online 24/7. This is the only way to make sure your scheduled jobs and alerts actually trigger. Using a VPS also keeps your main computer safe and isolated.

Hostinger provides a one-click template that handles the hard part of the installation. For most users, the KVM-2 plan is the best choice. It gives you 2 CPU cores and 8GB of RAM, which is more than enough to keep the agent snappy. a 12-month billing cycle usually offers the best balance of savings and a free domain.

During the setup, you must create an admin username and password. These are used to log into the web terminal. Hostinger generates a strong password by default. Copy this and save it in a secure place immediately. If you lose these credentials, you will be locked out of your agent's backend.

Once the project deploys, head to the VPS dashboard and open the web terminal. You need to enter the project directory first. Use the command cd /docker/hermes-agent/[your_last_four_chars]. After that, get inside the running container with docker compose exec -it hermes-agent /bin/bash. This puts you inside the environment where Hermes actually lives.

Core Configuration: AI Model and Telegram Integration

Now it is time to give your agent a brain. Run the hermes setup command and choose the Quick Setup option. When asked for a provider, select OpenRouter. OpenRouter is a huge win because it gives you access to over 200 different AI models with one single API key. You can switch from a cheap model to a powerful one without changing your setup.

After signing up for OpenRouter, create an API key specifically for your agent. It is smart to set a spending limit, like $50, so you don't get a surprise bill. Once you have the key, go to the Credits section in OpenRouter and add funds. An API key with no credits is just a dead link. Paste the key into your terminal using Ctrl+Shift+V. The text won't appear for security reasons, but it is being entered.

Next, choose your AI model. GPT-4o mini is a great starting point because it is fast and cheap. Now, set up the messaging gateway. Select Telegram from the list. To make this work, you need to talk to @BotFather on Telegram. Tell BotFather you want a /newbot, give it a name and a username, and copy the Bot Token he gives you.

Paste that token into the terminal. Finally, you need your numeric User ID so the bot knows who you are. Message @userinfobot on Telegram to get this ID. Paste it into the terminal and confirm that your ID should be the home channel. Now, Hermes knows exactly where to send your reports and notifications.

Understanding and Training Your Agent's Intelligence

Hermes uses skills, which are essentially recipes for tasks. It comes with 75 pre-installed skills. You can see them by asking, "Tell me each of the 75 skills that you currently have installed." These cover most basic needs, but the real power is in creating your own.

You can teach Hermes new workflows through a simple conversation. For example, ask it to research trending AI tools for a video. Once it finishes, tell it: "Create a new skill based on your approach and call it YouTube video research." Hermes will save that process. Next time you need research, it won't guess; it will follow the exact steps it just created.

Memory is the second pillar of the agent. Unlike ChatGPT, Hermes doesn't reset. It stores facts about you in a permanent database. Try saying, "Remember that I like my responses to be short and direct." You can check what it knows by asking it to show your memory. If it remembers something wrong, just tell it to forget that specific detail.

If you want more than just a list of facts, you can upgrade to Honcho. While built-in memory stores data, Honcho recognizes patterns in how you work. It thinks about your preferences between conversations. To switch, run hermes memory setup in the terminal and follow the prompts to connect a Honcho account.

Automating Workflows with Scheduled Jobs and Cost Control

This is where Hermes becomes a true assistant. You can schedule tasks using normal language. For example, tell it: "Set up a weekly job every Monday at 9:00 a.m. that researches the top three trending AI tools using the YouTube video research skill." Now, every Monday, you will wake up to a fresh report in your Telegram chat without typing a single word.

You can keep track of all your autonomous tasks by asking, "Show me a list of all the current jobs." This list shows the job ID, the schedule, and when the next run will happen. You can pause or delete these jobs anytime through the chat.

Costs can add up if you use the most powerful models for everything. Use a rule of thumb: use expensive models for hard thinking and cheap models for routine tasks. You can change your default model using the /model command.

You can also assign specific models to specific jobs. If your weekly research needs more "horsepower," tell Hermes: "I want the weekly AI tools research job to run on Opus 4.6." Now, your daily chats stay cheap with GPT-4o mini, but your high-value reports use a top-tier model.

Next Steps and VPS Management Overview

To keep the bot online 24/7, you must start the gateway. In the terminal, run hermes gateway install and then hermes gateway run. This separates the bot from the terminal, allowing it to live in the cloud. You can now close the web terminal and do everything through Telegram.

When you first message your bot, give it context. Send a message like: "You are running on a Hostinger VPS in a Docker container. Tell me what tools and skills you have." This helps the agent understand its environment immediately.

For security, remember that Hermes has four guardrails:

• Pairing Codes: Only users you approve can talk to the bot.
• Docker Isolation: The bot cannot access your host server.
• Command Approvals: It will ask you before deleting files.
• Skill Scanner: It checks new skills for bad code.

Beyond the basics, Hermes supports voice transcription and text-to-speech. You can also add it to Discord or WhatsApp via the gateway setup. If you want to connect it to GitHub, look into MCP support using the hermes mcp add command.

Finally, use your Hostinger dashboard for maintenance. The Docker manager lets you change admin passwords. The backups tab is the most important part. Always create a snapshot before you make a major change. If you break something, you can restore the server to a working version in one click.

Conclusion

Setting up a Hermes agent moves you from using a tool to owning an employee. You now have a 24/7 assistant that remembers your preferences and handles repetitive research on a schedule. By combining a VPS, OpenRouter, and Telegram, you've built a system that actually gets smarter the more you use it.

The key to success is to start small. Don't connect your bank account or main email on day one. Use exclusive accounts and build trust with the agent. Start with a few simple skills and one or two weekly jobs. As the agent proves its reliability, you can give it more complex tasks. Your AI partner is now ready to evolve alongside your workflow.