Setting Up a Personal AI Agent with Hermes — Maxxoto

I’ve been running Hermes Agent on my homelab for a few months now. Here’s what I’ve learned about setting it up, configuring it for real-world use, and connecting it to Telegram.

Why Hermes?

Most AI coding tools are closed-source, tied to a specific IDE, or require sending your code to a third-party API. Hermes is different:

Open source — MIT license, runs entirely on your hardware
Platform-agnostic — CLI, Telegram, Discord, or API server
Provider-agnostic — swap between OpenRouter, Anthropic, OpenAI, local models
Extensible — skills, plugins, MCP servers, custom tools

Installation

Getting started is straightforward:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

This sets up everything in ~/.hermes/ — config, skills, and a Python virtual environment.

Configuration

After installation, run the setup wizard:

hermes setup

You’ll be prompted to choose:

A model provider (I use OpenRouter)
A terminal backend (local is fine for a homelab)
Gateway platforms (Telegram, if you want mobile access)

Telegram Setup

Connecting Hermes to Telegram gives you an AI assistant in your pocket:

Create a bot via @BotFather and get your token
Add the token to ~/.hermes/.env:

TELEGRAM_BOT_TOKEN=123456...

Start the gateway:

hermes gateway start

That’s it. You can now message your bot and get responses from Hermes.

Skills & Memory

Where Hermes really shines is its skill system. Skills are reusable procedures that the agent loads into context when relevant. You can create your own:

hermes skills create my-deploy-workflow

Memory persists across sessions — Hermes remembers your preferences, project conventions, and past decisions. No more repeating yourself every conversation.

Running on a Homelab

My setup uses a Linux server on my local network. Key tips:

Run the gateway as a systemd service for auto-restart
Enable systemd user linger so it survives SSH logout
Use a custom Omniroute provider for local model inference

What’s Next

I’m currently exploring multi-agent orchestration — spawning specialized sub-agents for code review, research, and system administration tasks. The goal is a team of AI agents that work together on complex engineering workflows.