Easiest Step-by-step Hermes agent guide - Setup + Workflow

Step-by-step Hermes Agent setup guide for 2026. Learn how to install, connect Telegram, choose a model, build skills, and create automated workflows that run 24/7 on your own infrastructure.

Hermes Agent is an open-source, self-improving AI agent built by Nous Research. It lives in your terminal or cloud server, connects to 200+ AI models, and can be reached from Telegram, WhatsApp, Discord, Slack, or Signal - 24 hours a day, 7 days a week.

What makes it genuinely different from other agents is its self-learning loop: every 15 tool calls, it pauses, reviews what worked in the session, and writes a reusable skill file that makes it better at that task next time.

PART 1: BEFORE YOU START

System Requirements

Hermes runs on Linux, macOS, or WSL2 on Windows. Native Windows is not officially supported for the full setup, so Windows users should install WSL2 first.

Table 1.0

Table 1.0

What you need before installing

• An API key from a model provider.

• A Telegram account (or Discord/Slack) to interact with your agent.

• A VPS or local machine to host it.

PART 2: CHOOSE WHERE TO RUN IT

There are three ways to run Hermes Agent.

Option 1: VPS (Recommended) The best long-term option. Your agent runs 24/7 in the cloud even when your laptop is off. Hostinger VPS with the one-click Hermes template is the easiest entry point and costs roughly $4–5/month.

Option 2: Local machine Run it directly on your Mac or Linux machine. Free and private. Requires Ollama installed locally. Good for testing, but your agent goes offline when your computer does.

Option 3: Hosted platforms Services like Minimax offer a “Max Hermes” hosted option with minimal setup, but with limited integrations compared to a full VPS install.

PART 3: INSTALLATION

Step 1: Install Hermes

Run this command in your terminal:

bashcurl -sSL https://hermes-agent.nousresearch.com/install.sh | bash

That is it. Hermes installs in under 60 seconds on Linux or macOS.

Step 2: Get an API key

You need an LLM model to power your agent. The most popular options are:

• OpenRouter — recommended for beginners. Gives access to 200+ models including Claude, GPT, DeepSeek, and Gemma from one API key. Sign up at openrouter.ai.

• Anthropic API — if you want to run Claude directly.

• Ollama (free, local) — run models like Gemma 4 or Llama 3 completely free and offline on your own machine.

For most people, OpenRouter + a cheap model like Qwen 3.6 is the best starting point — capable, low cost, and fast.

Step 3: Run the setup wizard

After installing, start the Hermes setup wizard:

bash hermes setup

The wizard will walk you through:

• Entering your API key.

• Selecting your default model.

• Configuring memory settings.

• Setting up your messaging gateway (Telegram, Discord, etc.).

Step 4: Create a Telegram bot

Telegram is the cleanest way to interact with Hermes from any device.

• Open Telegram and search for @BotFather.

• Type /newbot and follow the prompts.

• Name your bot anything you want.

• Copy the Bot Token you receive.

• Paste it into the Hermes setup wizard when prompted.

Once paired, you can message your agent from your phone anywhere in the world.

Step 5: Start the gateway

bash hermes gateway start

This activates the messaging layer so your agent can receive and respond to messages via Telegram or any other connected platform.

Step 6: Send your first message

Open Telegram, find your bot, and type anything. Hermes should respond. Your agent is now live.

PART 4: UNDERSTANDING THE SYSTEM

Before building workflows, it helps to understand the four core layers of Hermes.

1. The Agent Core

The LLM brain that processes your messages, reasons through problems, and decides what tools to call.

2. The Memory System

Hermes has two layers of memory:

• Session memory — remembers what happened in the current conversation.

• Persistent memory — remembers facts, preferences, and patterns across sessions using FTS5 full-text search with LLM summarization.

You can optionally upgrade to Honcho for a more advanced user modelling layer that builds a deep profile of how you think and work over time.

3. The Skills System

This is Hermes’ most unique feature. Skills are reusable workflow files stored in ~/.hermes/skills/.

Every 15 tool calls, Hermes pauses and writes or updates a skill based on what worked in that session. The next time you do a similar task, it already knows the most efficient path. Skills are written as plain Markdown files you can read, edit, or delete at any time.

4. The Gateway

The communication layer that connects Hermes to Telegram, Discord, Slack, WhatsApp, Signal, Teams, or your terminal.

PART 5: CORE WORKFLOWS

Workflow 1: Daily Briefing (Automated)

Ask Hermes to send you a morning briefing every day.

Setup prompt in Telegram:

“Every morning at 7am, send me a briefing with: today’s weather, 3 top AI news headlines, my top 3 tasks for the day, and any messages I need to follow up on.”

Hermes will create a cron job, schedule the task, and deliver it automatically every day without being prompted again.

Workflow 2: Competitive Intelligence Monitor

This is the workflow described by one power user who ran the same prompt every Monday for six weeks. By week six, Hermes had rewritten the underlying skill four times and was surfacing competitor patterns they had not caught manually.

Setup prompt:

“Every Monday at 9am, search for news and updates about [Competitor A] and [Competitor B]. Summarize key moves, product changes, and pricing updates. Save a log in my memory and send me the briefing on Telegram.”

Workflow 3: Research Summarizer

Use Hermes to summarize articles, papers, or long documents on command.

How to use:

“Read this URL and give me: 5 key points, 2 surprising insights, and 3 questions I should be asking about this topic.”

After a few sessions, Hermes builds a skill around your summarization style and gets faster and more relevant each time.

Workflow 4: Writing Assistant

Use Hermes to draft, edit, or rewrite content directly from Telegram.

How to use:

“Take this rough idea and turn it into a LinkedIn post. My style is direct, no fluff, under 150 words, with a strong hook.”

After a few sessions, Hermes learns your style and stops asking clarifying questions. It just writes the way you like.

Workflow 5: Custom Skill Creation

You can also create custom skills manually by writing a SKILL.md file.

Drop the file in ~/.hermes/skills/ and it loads automatically. A simple skill file looks like this:

**# Skill: Weekly Review

Trigger

When I ask for a weekly review

Steps

1. Ask for my wins, losses, and key decisions from the week.
2. Summarize them into a structured note.
3. Suggest 3 priorities for next week.
4. Save the note to memory with today’s date.

Output format

Structured Markdown with clear headers.**

PART 6: MODEL SELECTION

One of Hermes’ best features is being fully model-agnostic. You can assign different models to different jobs.

Table 2.0

Table 2.0

PART 7: SECURITY BASICS

Hermes gives you full control, which means you also carry the responsibility of keeping it safe.

• Do not expose your gateway port publicly without authentication.

• Use command approval mode so Hermes asks before executing system commands.

• Review your skills folder periodically to audit what your agent has learned.

• Set spending limits on OpenRouter to prevent unexpected API costs.

• Use container isolation (Docker mode) if you want an extra security layer for code execution.

PART 8: WHAT ELSE IS POSSIBLE

Once you have the basics running, Hermes can go much further.

• Voice mode — send voice messages from Telegram, get voice replies back with text-to-speech.

• MCP integration — connect Hermes to Model Context Protocol servers for additional tools.

• Multi-agent profiles — create multiple agent personalities with different skills and models using hermes profile create.

• Web search — add Firecrawl or a web search API so Hermes can research the live web.

• Image generation — connect Google Imagen or a compatible API for image tasks.

• Multi-platform delivery — run one agent across Telegram, Slack, Discord, and WhatsApp simultaneously.

Hermes Agent is not powerful on day one. It is powerful on day thirty. Atleast I learned this hard way!

The reason is the compounding loop: every task you repeat teaches Hermes something about how you work. Every skill it writes makes the next session faster. Every memory it stores gives future sessions better context.

Set it up once. Use it consistently. Watch it become the most useful assistant you have ever worked with.

I enjoyed writing this article. Just incase if you looking for a FREE HERMES AGENT GUIDE, don’t forget to checkout my Gumroad Shop👇

**https://passiveincomehabit.gumroad.com/ ** Also this is my newsletter: https://passiveincomehabit.beehiiv.com/

I am Kanika (@KanikaBK), specializing in AI tools, emerging trends, and niche applications.

Follow for in-depth analyses, strategic insights, and professional updates to elevate your AI knowledge.