DEV Community

Cover image for How to Deploy Hermes' Self-Improving AI Agent
DigitalOcean profile image haimantika mitra
haimantika mitra for DigitalOcean

Posted on with Anish Walia • Originally published at digitalocean.com

How to Deploy Hermes' Self-Improving AI Agent

#hermes #ai #agents #tutorial

Key takeaways

  • You host Hermes on a DigitalOcean Droplet (Ubuntu 24.04, at least 2 vCPUs and 4 GB RAM), install with the official script, and run hermes setup for your LLM provider.
  • The Telegram bot uses hermes gateway setup and a persistent gateway service so Hermes stays reachable after you close SSH.
  • Skills are Markdown files under ~/.hermes/skills/. MCP servers go in Hermes config and expose tools such as grocery search and cart actions.
  • HTTP MCP with OAuth on a headless server uses the URL Hermes prints plus an SSH tunnel from your laptop browser to the Droplet loopback port.
  • The grocery walkthrough uses Swiggy Instamart where available; you swap in another MCP URL for your region or stack.

2026 is the year of agents doing the task for us rather than just guiding us to. We saw the rise of OpenClaw and now we have Hermes agent. It is an open-source, self-improving AI agent built by Nous Research. It has a built-in learning loop that creates skills from experience, improves them during use, nudges itself to persist knowledge, and builds a deepening model of who you are across sessions.

In this tutorial, you will deploy Hermes on a DigitalOcean Droplet, connect it to Telegram, and extend it with a custom skill. As a practical example, you will see how to build a grocery tracking agent that monitors daily consumption, alerts you when stock is low, and places orders automatically through a grocery delivery service.

The grocery example uses Swiggy Instamart, which is available in India. But the approach works with any MCP-compatible service, such as a local grocery API, a task manager, a calendar, or a home automation system. The pattern is the same no matter what you connect.

Prerequisites

Before you begin, you will need:

  • A DigitalOcean account. If you do not have one, sign up here.
  • A DigitalOcean Droplet running Ubuntu 24.04 with at least 2 vCPUs and 4 GB RAM.
  • A Telegram account.
  • API key from an LLM provider. Hermes supports Anthropic, OpenAI, OpenRouter, and others.

What are we building

Hermes is model-agnostic and platform-agnostic. It can connect to any tool that supports the Model Context Protocol (MCP), and it can send and receive messages on Telegram, WhatsApp, Discord, and Slack. You can find more information on their documentation.

The architecture in this tutorial looks like this:

Architecture

Let’s get started building:

Step 1. Create the Droplet

Sign in to your DigitalOcean account and create a new Droplet.

On the creation page, select:

  • Region: The one closest to you
  • Image: Ubuntu 24.04 LTS
  • Plan: Basic, with at least 2 vCPUs and 4 GB RAM (the s-2vcpu-4gb size)
  • Authentication: SSH Key

If you need help adding an SSH key, follow the DigitalOcean SSH key guide.

Once the Droplet is created, SSH into it:

ssh root@YOUR_DROPLET_IP
Enter fullscreen mode Exit fullscreen mode

Update the system before installing anything:

apt update && apt upgrade -y
Enter fullscreen mode Exit fullscreen mode

Step 2. Install Hermes Agent

Hermes provides an install script that handles all dependencies including Python, uv, and the hermes binary itself.

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
Enter fullscreen mode Exit fullscreen mode

Once it finishes, reload your shell so the hermes command is available:

source ~/.bashrc
Enter fullscreen mode Exit fullscreen mode

Run the setup wizard:

hermes setup
Enter fullscreen mode Exit fullscreen mode

The wizard will ask you to choose an LLM provider and enter your API key. Select your provider, paste your key, and Hermes will save it to ~/.hermes/.env.

Verify the installation:

hermes --version
Enter fullscreen mode Exit fullscreen mode

Step 3. Connect Hermes to Telegram

Hermes connects to Telegram through a bot. Run the gateway setup:

hermes gateway setup
Enter fullscreen mode Exit fullscreen mode

When it asks which platform to use, select Telegram. Hermes will walk you through creating a bot with Telegram's BotFather. Follow the steps in your terminal.

Once setup is complete, start the gateway:

hermes gateway start
Enter fullscreen mode Exit fullscreen mode

To keep it running after you disconnect from the Droplet, enable it as a system service:

systemctl enable hermes-gateway
systemctl start hermes-gateway
Enter fullscreen mode Exit fullscreen mode

Open Telegram, find the bot you just created, and send it a message like "hello." If it responds, Hermes is connected and ready.

You can now talk to Hermes from anywhere on your phone. Ask it to check the weather, set a reminder, search the web, or run a command on your Droplet. It handles all of this out of the box.

Step 4. Understand skills and MCP servers

Before building the automation example, let's understand two core Hermes concepts:

Skills - These are Markdown files that teach Hermes how to handle specific tasks. You write a skill file describing what the task is, what triggers it, and what steps to follow. Hermes reads all skill files in ~/.hermes/skills/ at startup and uses them to handle relevant requests.

MCP Servers - It gives Hermes access to external services. MCP (Model Context Protocol) is an open standard that lets AI agents communicate with APIs in a structured way. If a service publishes an MCP server, Hermes can search its products, manage carts, read calendars, create tasks, and more. You add MCP servers to your Hermes config and Hermes uses them automatically when relevant.

Together, skills and MCP servers let you build automations that are specific to your life and the tools you use.

Step 5. Build real-world automation: Grocery tracking

Personally, I am very bad at keeping a tab of groceries and often find myself out of essential stock just when I am about to cook. So I wanted to automate this process for me. I eat similar food everyday and also measure my calories. I used all of this information to automate grocery shopping for me.

The goal of Hermes was to track my daily grocery consumption, alert me on Telegram when something is running low, and place an order through a grocery delivery service when I say yes.

This example uses the Swiggy Instamart MCP server, which is available in India. If you are in a different country, you can swap it out for any MCP-compatible grocery or delivery service. The skill logic stays exactly the same.

Creating the skill file

Create a directory for the skill:

mkdir -p ~/.hermes/skills/grocery
Enter fullscreen mode Exit fullscreen mode

Create the skill file:

nano ~/.hermes/skills/grocery/grocery_tracker.md
Enter fullscreen mode Exit fullscreen mode

Paste the following and edit the groceries: section to match your actual diet and quantities:

# Grocery Auto-Order Skill

### My Daily Grocery List

groceries:
  - name: "Milk"
    unit: "ml"
    daily_consumption: 500
    reorder_quantity: 2000
    alert_threshold_days: 3
    search_query: "fresh milk 1 litre"

  - name: "Eggs"
    unit: "pieces"
    daily_consumption: 2
    reorder_quantity: 12
    alert_threshold_days: 3
    search_query: "eggs 12 pack"

  - name: "Oats"
    unit: "gm"
    daily_consumption: 60
    reorder_quantity: 1000
    alert_threshold_days: 7
    search_query: "rolled oats 1kg"

  # Add your own items following the same format

### Instructions for Hermes

#### Daily check
Read ~/.hermes/grocery_inventory.json. Subtract each item's
daily_consumption from its current quantity. For any item where
quantity / daily_consumption is less than or equal to
alert_threshold_days, send a Telegram alert listing the low
items and ask if the user wants to place an order.

#### On YES
Search each low item using its search_query at the user's saved
delivery address. Add reorder_quantity of each to cart. Send a
cart summary via Telegram with prices and total. Wait for CONFIRM.

#### On CONFIRM
Place the order. Update grocery_inventory.json with the restocked
quantities. Send a delivery confirmation message.
Enter fullscreen mode Exit fullscreen mode

Save with Ctrl+O, then Enter, then exit with Ctrl+X.

Creating the inventory file

The inventory file tracks how much of each item you have at home right now.

nano ~/.hermes/grocery_inventory.json
Enter fullscreen mode Exit fullscreen mode 
{
  "last_updated": "2026-05-07",
  "delivery_address": "YOUR FULL ADDRESS HERE",
  "snooze_until": null,
  "items": {
    "Milk": { "quantity": 1000, "unit": "ml" },
    "Eggs": { "quantity": 6, "unit": "pieces" },
    "Oats": { "quantity": 300, "unit": "gm" }
  }
}
Enter fullscreen mode Exit fullscreen mode

Replace the quantities with what you actually have at home. Save and exit.

Connect to an MCP server

Add the grocery service MCP to your Hermes config:

hermes config edit
Enter fullscreen mode Exit fullscreen mode

Scroll to the bottom of the file and add your MCP server. For Swiggy Instamart:

mcp_servers:
  swiggy-instamart:
    url: "https://mcp.swiggy.com/im"
    auth: oauth
Enter fullscreen mode Exit fullscreen mode

For any other MCP-compatible service, replace the name and URL with the values from that service's documentation.

Save and exit, then verify it appears:

hermes mcp list
Enter fullscreen mode Exit fullscreen mode

Step 6. Authenticate with your MCP server

Most MCP servers require OAuth authentication. Because your Droplet is headless, the OAuth callback needs to reach your browser through an SSH tunnel.

Run the login command on your Droplet:

hermes mcp login swiggy-instamart
Enter fullscreen mode Exit fullscreen mode

Hermes will print a URL containing a port number in the redirect URI, like http://127.0.0.1:45123/callback. Note that port number.

On your local machine, open a new terminal and run the SSH tunnel using that port:

ssh -i ~/.ssh/id_ed25519 -L 45123:127.0.0.1:45123 root@YOUR_DROPLET_IP
Enter fullscreen mode Exit fullscreen mode

Keep that terminal open, then immediately open the URL from your Droplet in your browser and complete the login. You have about 30 seconds before the token expires, so have both terminals ready before you start.

When authentication succeeds, your Droplet terminal will show:

✓ Authenticated — tools available
Enter fullscreen mode Exit fullscreen mode

Telling Hermes Your Current Stock

Start Hermes and initialize the inventory:

hermes
Enter fullscreen mode Exit fullscreen mode

Type:

Set up my grocery tracker. Ask me for current stock levels for each item in the grocery_tracker skill.
Enter fullscreen mode Exit fullscreen mode

Hermes will ask about each item one by one. Answer with what you currently have at home and it will update grocery_inventory.json automatically. Once you have given the update, you will receive a message similar to this:

image2

Setting up the daily alert

Still inside Hermes, set up the cron job:

Add a daily cron at 8am: check my grocery inventory using the grocery_tracker skill, subtract daily consumption, and send me a Telegram message if anything is running low.
Enter fullscreen mode Exit fullscreen mode

Hermes will schedule this and confirm. From now on, every morning it checks your stock and messages you on Telegram if you need to reorder. This is how it looks:

image3

Step 7. Test the full flow

Send a test message to your Telegram bot:

Check my groceries and tell me what's running low.
Enter fullscreen mode Exit fullscreen mode

You should receive a Telegram message like this:

image4

Reply YES. Hermes searches your connected grocery service, builds a cart, and sends a summary.

Once everything is set up, you can manage your agent entirely from Telegram:

Message What Hermes does
Check my groceries Shows all items and days of stock remaining
I bought 12 eggs Updates egg stock in the inventory file
Order groceries now Skips the check and goes straight to ordering
What's running low? Lists items near their alert threshold
SKIP Snoozes today's alert until tomorrow

Conclusion

You now have a self-hosted AI agent running on DigitalOcean that works for you around the clock. Hermes connects to Telegram and more messaging platforms so you can reach it from anywhere, and skills plus MCP servers let you extend it to handle almost anything.

The grocery automation you built in this tutorial is a starting point. The pattern is reusable: write a skill file that describes the task, connect an MCP server that gives Hermes access to the right service, and set a cron job to trigger it automatically. The grocery automation is one example. The same pattern works for any repetitive task in your life. A few ideas to get you started:

Bill reminders- Create a skill that tracks recurring payments, calculates due dates, and alerts you three days before each bill is due.

Health tracking- Log your workouts or meals via Telegram and have Hermes summarize your week every Sunday.

Home automation- Connect Hermes to a smart home MCP server and have it adjust lights, thermostats, or appliances based on a schedule or your location.

Related links

Top comments (0)