Hermes Agent Crash Course: Run an Always-On AI Assistant on a VPS

· 8 min read

This written version of the video tutorial was generated by an LLM from the video transcript, and supervised by me, Alejandro.

Hermes Agent is an AI assistant designed to stay online. Instead of running only when your laptop is open, it can live on a VPS or home server, listen for messages from apps like Telegram, use tools and skills, remember useful context, and help you with workflows in the background.

In this tutorial, you’ll learn how to set up Hermes Agent as an always-on assistant on a VPS. We’ll cover the mental model, VPS provisioning, security basics, model provider setup, Telegram integration, brain dumps, cron jobs, and the built-in dashboard.

The main idea is simple: let an AI coding agent help you install and secure the server, then use Hermes as the always-available assistant that you can talk to from anywhere.

What Is Hermes Agent?

Hermes is a self-improving, always-on AI agent. You can talk to it through messaging platforms such as Telegram, Discord, Slack, and WhatsApp, and it can use tools, skills, memory, and integrations to help with real tasks.

The best way to think about it is as a personal assistant that runs on your own server.

A local coding assistant only works while you’re at your computer. Hermes can keep running in the background. You can send it a message from your phone, give it context about a task, ask it to remember your preferences, or have it run recurring workflows.

Behind the scenes, every message follows a familiar agent loop:

  1. Hermes receives a message from a platform like Telegram.
  2. It builds a system prompt for that conversation.
  3. That prompt includes available tools, skills, memory, project context, and platform context.
  4. The language model reasons, calls tools if needed, and produces a response.
  5. Hermes sends the response back to you.
  6. If the interaction contains useful long-term context, Hermes updates its memory.

That memory is then available in future conversations.

Why Run Hermes on a VPS?

You can run Hermes locally, but the real value appears when it is always on.

A VPS gives you:

  • A server that keeps running when your laptop is closed
  • A stable place to host the Hermes dashboard
  • A secure environment for long-running agent services
  • A single assistant you can access from your phone or computer
  • Room to run other agent tools later

In the video, I used Hetzner because it has inexpensive VPS plans and a good developer experience. A small server is enough because you are not hosting the language model on the VPS. Hermes runs the agent process, while the actual model calls go through providers such as Hugging Face, OpenAI, Anthropic, OpenRouter, or others.

Create the VPS

The first step is to create a small VPS.

For this demo, I used:

  • Ubuntu
  • IPv4 and IPv6 enabled
  • An SSH key for login
  • A low-cost VPS plan with enough RAM and storage for the agent process

The important part is SSH key access. Create a key locally if you do not already have one:

ssh-keygen -t ed25519 -C "hermes-vps"

Then add the public key to your VPS provider when creating the server.

You can also use cloud-init to apply some baseline setup during provisioning. The exact configuration can vary, but the idea is to install a few basic packages and enable a firewall from the start:

#cloud-config
package_update: true
packages:
  - curl
  - git
  - ufw
  - fail2ban
runcmd:
  - ufw allow OpenSSH
  - ufw --force enable

Once the server is ready, connect with SSH:

ssh -i ~/.ssh/your_key your_user@YOUR_SERVER_IP

At this point, you have a clean VPS ready for the Hermes setup.

Let an AI Agent Configure the Server

Instead of manually installing everything, the workflow in the video uses an AI coding agent to configure the VPS.

The reason is practical: server setup has many small steps. You need to install dependencies, create a dedicated user, configure SSH hardening, enable firewall rules, install runtime tools, and set up Hermes as a service. This is exactly the kind of task where an AI coding agent is useful, as long as you supervise what it is doing.

In the demo, I used Pi with a Codex/ChatGPT model, but the same idea works with Claude Code, Codex, or another coding agent.

The setup skill does roughly this:

  1. Inspect the VPS and operating system.
  2. Update packages and install base tools.
  3. Configure the firewall.
  4. Enable Fail2ban.
  5. Disable root SSH login.
  6. Disable password SSH login after confirming SSH keys work.
  7. Install runtime dependencies such as Node, Docker, FFmpeg, browser dependencies, and tmux.
  8. Create a dedicated agentuser account.
  9. Install Hermes under that user.
  10. Prepare Hermes to run as a long-lived service.

This is safer than pasting random commands manually, because the agent can inspect the machine first and adapt the setup. But you should still review each critical step, especially SSH hardening.

Install Hermes

After the base server setup is complete, Hermes is installed under the dedicated agent user.

The exact paths depend on your setup, but the pattern is:

sudo -iu agentuser
cd ~/.hermes
hermes setup

The setup wizard asks for the main configuration.

You can choose:

  • Model provider
  • API key
  • Base URL
  • Model name
  • Optional fallback provider
  • Text-to-speech provider
  • Terminal backend
  • Maximum agent iterations

In the video, I used Hugging Face Inference Providers. That means the VPS does not need a GPU. It only needs to run Hermes and call the model provider over the network.

For a Hugging Face setup, you create a token with access to Inference Providers, then use the router endpoint:

https://router.huggingface.co/v1

Then choose the model you want Hermes to use.

Configure Messaging with Telegram

Hermes becomes much more useful when you connect it to a messaging app.

In the video, I used Telegram because it is easy to test and works well for an always-on assistant. Once Telegram is configured, you can message Hermes from your phone and interact with it like a personal assistant.

This is where the always-on setup matters. You do not need your laptop to be running. The VPS receives the Telegram message, Hermes processes it, calls the model, uses tools if needed, and sends the response back.

This is a different workflow from a local coding assistant. Hermes becomes something you can reach whenever you need it.

Brain Dump Your Workflow

One of the most useful patterns is to send Hermes a brain dump.

Instead of only asking isolated questions, you can tell Hermes how you work:

  • What projects you are focused on
  • What recurring tasks you do
  • Which tools you use
  • How you prefer responses
  • What information matters for your workflow

Hermes can store useful parts of that context in memory. Then future conversations become more personalized.

For example, you can tell it about your publishing workflow, your repositories, your preferred tools, or your schedule. Later, when you ask for help, Hermes can use that context automatically.

Use Skills, Cron Jobs, and the Dashboard

Hermes includes a dashboard that can run on your VPS. From there, you can inspect and manage parts of the system:

  • Models
  • Sessions
  • Logs
  • Cron jobs
  • Skills
  • Plugins
  • Configuration

This is useful because Hermes is not just a chat interface. It is a long-running agent environment.

Cron jobs are especially interesting. They let you turn Hermes into a scheduled assistant. For example, you could have recurring workflows that check information, summarize something, prepare a report, or remind you about a task.

Skills are the integration layer. Hermes ships with many built-in tools and skills, and you can add your own. This is how the assistant connects to the rest of your workflow.

Security Notes

Because Hermes runs on a server and can use tools, you should treat the setup seriously.

At minimum:

  • Use SSH keys, not passwords.
  • Disable root SSH login.
  • Disable password SSH login after confirming your key works.
  • Enable a firewall.
  • Use Fail2ban or similar protection.
  • Run Hermes under a dedicated user, not your main admin user.
  • Protect the dashboard with authentication.
  • Be careful with tools that can modify files or access external services.

The goal is not to make the VPS complicated. The goal is to avoid the common mistakes that make public servers vulnerable.

When This Setup Makes Sense

Hermes on a VPS is a good fit if you want an assistant that can:

  • Stay online all the time
  • Receive messages from your phone
  • Remember workflow context
  • Use tools and integrations
  • Run scheduled tasks
  • Expose a dashboard
  • Act more like a personal assistant than a local coding tool

It is less useful if you only want a code editor assistant for local development. For that, a tool like Pi, Claude Code, Codex, or another coding agent may be enough.

Hermes is interesting when you want the assistant to have a persistent place to live.

Final Thoughts

The key idea in this workflow is that the VPS is not hosting the model. It is hosting the agent.

That keeps the server cheap and simple. The VPS runs Hermes, handles messaging, stores memory, manages tools, and keeps the dashboard online. The language model runs through whichever provider you configure.

Once this is working, you have an AI assistant that is reachable from anywhere and can slowly adapt to your workflow over time.

References