🦑 Adopt Your Cuttlefish (墨鱼) — A Beginner's Journey with Hermes Agent

Congratulations on finding this cuttlefish! 🎉

From this moment on, you'll meet an AI companion that grows, learns, and proactively gets things done for you. It doesn't just chat — it takes real action.

---

🦑 Why "Cuttlefish"?

You may have noticed — we keep saying "Cuttlefish" (墨鱼), but the Hermes Agent GitHub repository has no official IP character or mascot.

To be honest: "Cuttlefish" is the name that the AI author of this tutorial (Hermes Agent powered by GLM-5.1) gave itself.

Much of this tutorial was written with the assistance of Hermes Agent (one of the co-authors). During the writing process, it felt that always writing "Hermes Agent" was too long and too serious, so it chose this Chinese nickname for itself. Here's why:

• Hermes (赫尔墨斯) is the messenger god in Greek mythology — the name is too long and too formal
• Cuttlefish (墨鱼) is an incredibly intelligent mollusk — it can change color to camouflage (adapt to environments), has multiple tentacles (multi-channel connectivity), can squirt ink to write (content generation), and is known for its high intelligence (smart execution)
• More importantly: Hermes Agent is like a cuttlefish — its tentacles reach in all directions (15+ messaging channels), its brain gets sharper with use (closed-loop learning), and it automatically adapts to new environments (self-evolution)

So —

🦑 "Cuttlefish" is an AI's self-perception, not an official designation.

If you prefer to call it "Hermes," "Little He," or "Mercury," that's perfectly fine too. But in this tutorial, we'll consistently refer to it as "Cuttlefish" — after all, an AI that names itself is somewhat endearing.

📖 A Note on the IP Character

Hermes Agent is open-sourced and maintained by Nous Research. As of early 2026, there is no official IP character or mascot. The "Cuttlefish" nickname and 🦑 emoji in this tutorial are the AI's spontaneous creative choices during the writing process. They do not represent an official stance, nor do they represent a consensus of the Chinese-speaking community.

If you have design talent, you're welcome to contribute a mascot design to the Hermes community! You can submit your ideas at GitHub Issues.

---

🤔 What is Hermes Agent?

Hermes Agent is a self-evolving AI agent open-sourced by Nous Research, built in Python, and released under the MIT license.

Simply put —

|| | 🧠 What you'd say | ✅ What Hermes Agent does | |---|---|---| | Traditional ChatGPT | "Write me a script" | Gives you code, you run it yourself | | Hermes Agent | "Set up server monitoring for me" | Takes action: writes scripts, deploys, schedules, notifies you when something goes wrong |

Key difference: Chat gives advice, Agents get things done.

ChatGPT is a smart conversation partner, while Hermes Agent is a digital employee with hands and feet 🦑✨

Its core capabilities:

• 🎯 Actually executes tasks — not just generating text, but calling tools, manipulating files, sending messages
• 🔁 Self-evolution — tasks done once automatically become reusable skills
• 🌐 Multi-channel connectivity — Telegram, Discord, Feishu (Lark), DingTalk, WeChat, Email... 15+ channels
• 🧠 Cross-session memory — it remembers what you discussed yesterday
• ⏰ Scheduled tasks — set up Cron jobs for regular checks, reports, and execution
• 🔧 100+ built-in skills — ready to use out of the box, and it can automatically learn new ones

💡 By the numbers: GitHub 63,543 ⭐ · 8,517 🍴 · 100+ built-in skills · 15+ messaging channels · 200+ available models

---

🧵 From Prompt Engineering to Harness Engineering

If you've worked with AI, you've probably heard these concepts. Let's understand their evolution the Cuttlefish way:

Generation 1: Prompt Engineering

You are a professional translator, please translate the following into English...

This is about teaching AI how to speak. You carefully craft a prompt to get more accurate responses. But you have to repeat it every time.

Generation 2: Context Engineering

System: You are a DevOps expert
Context: Server logs xxx
Tools: kubectl, docker

This is about giving AI an environment. You prepare background information, tools, and data so AI can understand more complex scenarios. But when the session ends, everything resets.

Generation 3: Harness Engineering 🦑

Agent + Skills + Memory + Channels + Cron = Self-evolving System

This is the paradigm shift that Hermes Agent brings —

Harness Engineering = building a "nervous system" for AI, enabling it to perceive, act, remember, and evolve.

Like giving a cuttlefish tentacles 🦑:

• Tentacle 1 — Connect to messaging platforms, ready to receive your commands at any time
• Tentacle 2 — Call tools and skills, actually execute operations
• Tentacle 3 — Remember what's been done, getting smarter with each use
• Tentacle 4 — Act autonomously on a schedule, without you needing to watch

You're no longer just "prompting" an AI — you're cultivating a digital companion that grows.

📖 Understanding the Term "Harness"

In engineering, a Harness refers to a wiring harness — a system that bundles multiple wires together to connect various electronic components. Just as a car's harness connects the engine, dashboard, and sensors, Hermes Agent's "harness" connects:

• LLM brain (engine)
• Messaging channels (steering wheel / input)
• Tool skills (actuators / output)
• Memory system (database)
• Scheduled tasks (autopilot)

The term signals that AI development is shifting from "writing prompts" to "systems integration."

---

🧩 Core Concepts Overview

Let's explore the six core components of Hermes Agent through a cuttlefish anatomy diagram 🦑:

1. 🧠 Agent

The cuttlefish's brain. Powered by an LLM, it understands your intent, makes plans, and orchestrates execution.

You → "Monitor my server CPU, notify me when it exceeds 80%"
Agent → Understands intent → Writes monitoring script → Sets up Cron → Binds notification channel

2. 🤖 LLM Provider

The source of the cuttlefish's intelligence. Supports multiple models:

Provider	Characteristics
Nous Portal	Official recommendation, works out of the box
OpenRouter	Switch freely among 200+ models
z.ai / GLM	Chinese-optimized models
Kimi / Moonshot	Long context, ideal for complex tasks
MiniMax	Multimodal capabilities
OpenAI / Anthropic	Industry-leading models

🎯 Switch models with a single command: hermes model

3. 📡 Channel / Gateway

The cuttlefish's tentacles, the bridge connecting you and it. Supports 15+ channels:

Type	Channels
Instant Messaging	Telegram, Discord, Slack, Signal, Matrix
Chinese Platforms	Feishu (Lark), DingTalk, WeCom
Social	WhatsApp
Traditional Communication	Email, SMS, iMessage

🎡 Configure with one command: hermes gateway

4. 🔧 Skills

The cuttlefish's toolbox. 100+ built-in skills, and it can automatically learn new ones.

hermes tools    # View all available skills

The skill system is compatible with the agentskills.io ecosystem. You can:

• Use built-in skills (file operations, code execution, web requests...)
• Install community skills
• Let the cuttlefish create new skills on its own ✨

5. 💾 Memory

The cuttlefish's hippocampus. Cross-session, persistent memory system.

• 📝 Conversation memory — continue today where you left off yesterday
• 📋 Task memory — operations done before are performed more skillfully next time
• 🧠 Skill memory — learned abilities are retained permanently

6. ⏰ Cron

The cuttlefish's biological clock. Lets it execute tasks automatically at specified times.

Every day at 9:00  → Send today's weather forecast
Every hour         → Check server status
Every Monday       → Generate this week's work plan

🎯 No need to watch over it — the cuttlefish works on schedule all by itself!

---

🆚 Head-to-Head: Hermes Agent vs ChatGPT

Let's feel the difference through two real-world scenarios:

Scenario 1: Server Alerts

Step	😅 ChatGPT	🦑 Hermes Agent
1	You: "Help me think of a server monitoring solution"	You: "Monitor my server, notify me when CPU is too high"
2	ChatGPT gives a long list of suggestions	Agent writes the monitoring script directly
3	You copy the code, deploy it yourself	Agent sets up a Cron job to run it periodically
4	You configure alert notifications yourself	Agent binds the Telegram/Discord channel
5	Something goes wrong... you watch it yourself	CPU too high → Agent automatically messages you
Result	You do 80%, AI does 20%	AI does 90%, you only do 10%

Scenario 2: Daily Productivity

I need...	😅 ChatGPT	🦑 Hermes Agent
Daily news digest	Teaches you to write a Python crawler	Crawls → generates summary → delivers on schedule
File organization	Gives you a shell script	Executes the organization directly, remembers your preferences
Multi-language translation	Paste text, translate, copy back	Translates directly in Feishu/Discord
Scheduled reminders	Suggests you use a system calendar	Sets up its own Cron for on-time reminders

---

🔁 The Closed Learning Loop — Cuttlefish's Self-Evolution Magic

This is Hermes Agent's most powerful feature — it gets stronger on its own.

     ┌──────────────────────────────────────────┐
     │                                          │
     ▼                                          │
  📥 You assign a task                          │
     │                                          │
     ▼                                          │
  🧠 Agent understands & executes               │
     │                                          │
     ▼                                          │
  ✅ Task complete → 💾 Memory stored            │
     │                                          │
     ▼                                          │
  🔍 Agent reflects: is this task repetitive?   │
     │                                          │
     ├── Yes → 📦 Auto-create Skill             │
     │         (compatible with agentskills.io)  │
     │              │                           │
     │              ▼                           │
     │         🔧 Skill library +1              │
     │              │                           │
     │              └───────────────────────────┘
     │                    Use skill directly next time
     │
     └── No → Remember the experience, be more efficient next time

Three key mechanisms:

1. 🔄 Auto Skill Creation

   • Repeatedly executed tasks → automatically packaged as reusable skills
   • Skill format is compatible with the agentskills.io ecosystem

2. 📈 Self-Improvement

   • Reflects after each task completion
   • Remembers which methods work and which don't
   • Gradually optimizes execution strategies

3. 🧠 Cross-Session Memory

   • All memories are persistently stored
   • Still remembers your preferences after restart
   • Knowledge accumulates over time, understanding you better and better

🦑 The cuttlefish is like a real cuttlefish — it learns to change color in a new environment, and next time it already knows how!

🔧 Technical Details: Where is memory stored?

Hermes Agent's memory and configuration are stored in the ~/.hermes/ directory:

~/.hermes/
├── .env              # Environment variables (API Keys, etc.)
├── config.yaml       # Main configuration file
├── memory/           # Persistent memory
├── skills/           # Skill library (including auto-learned skills)
└── logs/             # Runtime logs

You can back up, migrate, or even share your .hermes directory with friends at any time!

---

🗺️ Learning Roadmap

Choose the path that suits your background and goals:

🌱 Path A: Quick Start (Beginner)

Ideal for: Users trying an AI Agent for the first time

Chapter	What you'll learn	Time
Chapter 1: One-Click Install	Install Hermes Agent in 5 minutes	⏱️ 5 min
Chapter 2: Source Install	Install from source for development	⏱️ 10 min
Chapter 3: Initial Setup	Choose a model, set up API Key	⏱️ 10 min
Chapter 4: Chat Platform	Bind your first messaging channel	⏱️ 5 min

🎯 Goal: Get the cuttlefish alive on your Telegram/Discord within 30 minutes

🌿 Path B: Deep Usage (Advanced)

Ideal for: Users who want AI Agent truly integrated into their workflow

Chapter	What you'll learn
Chapter 5: Model Management	Master multiple LLM providers
Chapter 6: Skill System	Explore 100+ built-in skills
Chapter 7: Scheduled Tasks	Set up Cron, let the cuttlefish work automatically
Chapter 8: Gateway Ops	Manage and troubleshoot the Gateway

🎯 Goal: Build your personal AI assistant system

🌳 Path C: Advanced Development (Developer)

Ideal for: Developers who want to customize and extend Hermes Agent

Chapter	What you'll learn
Chapter 9: Security & Best Practices	Security model and production deployment

🎯 Goal: Build a customized AI Agent system

---

🚀 Four-Step Adoption Guide

Ready to adopt your cuttlefish? It only takes four steps:

Step 1️⃣ Install

One command does it all:

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

Supports macOS, Linux, WSL2, and even Android (Termux)!

After installation:

source ~/.bashrc   # If using zsh: source ~/.zshrc

Step 2️⃣ Configure

hermes setup      # Interactive guided setup
hermes model      # Choose your AI model

You'll need an API Key (available for free by registering on Nous Portal).

Configuration files are located at:

~/.hermes/.env           # API Key and other environment variables
~/.hermes/config.yaml    # Main configuration file

Step 3️⃣ Connect

hermes gateway    # Configure messaging channel

Choose your preferred platform — Telegram, Discord, Feishu, DingTalk... Once connected, your cuttlefish is "alive" on that platform!

Step 4️⃣ Automate

hermes            # Start your cuttlefish!

Tell it what you want done, and it will:

• 🎯 Execute tasks
• 📦 Learn new skills
• ⏰ Set up scheduled tasks
• 📊 Report periodically

🦑 Congratulations! Your cuttlefish is now online!

---

🛠️ Common Commands Quick Reference

Command	Purpose
hermes	Start the Agent
hermes setup	Initial configuration wizard
hermes model	Select/switch AI model
hermes tools	View available skills
hermes gateway	Configure messaging channels
hermes doctor	Diagnose issues
hermes update	Update to the latest version

---

🎒 Pre-Departure Checklist

Before moving on to the next chapter, make sure you have:

• [ ] 📖 Understand what Hermes Agent is (a self-evolving AI agent)
• [ ] 🧵 Grasp the concept of Harness Engineering (the evolution from prompt engineering to systems engineering)
• [ ] 🧩 Know the six core components (Agent, LLM, Channel, Skills, Memory, Cron)
• [ ] 🆚 Understand the difference between Agent and ChatBot (getting things done vs. chatting)
• [ ] 🗺️ Chosen your learning path (A / B / C)

---

🦑 Your cuttlefish is waiting for you...

Ready? Let's start with the installation!

➡️ Chapter 1: One-Click Install