Get Started with Hermes in 3 Minutes: A Beginner’s Guide to Your Own AI Agent

Many AI assistants forget context after each session, forcing users to re‑teach preferences every time. Hermes Agent, developed by Nous Research and open‑sourced in February 2026, addresses this by continuously learning from interactions and retaining memory across sessions.

What Is Hermes?

Hermes Agent is a self‑evolving AI agent framework that aims to become a personalized assistant that grows with you. It automatically records effective methods as "Skills" after each task, remembers conversation history across terminal restarts, and supports multiple LLM back‑ends.

Core Capabilities

Auto‑learning : extracts reusable skills from completed tasks.

Cross‑session memory : retains project context, preferences, and decisions after the terminal is closed.

Multi‑model support : works with OpenRouter, OpenAI, Claude, Ollama, etc., allowing you to switch models freely.

Built‑in tools : file handling, web requests, data analysis are ready out of the box.

Local deployment : runs on a personal computer or small VPS without uploading data.

Security hardening : sandbox isolation and injection scanning are included by default.

v0.3.0 Highlights (released 2026‑03‑17)

Unified streaming output : token‑by‑token real‑time display.

Plugin architecture : drop Python files into ~/.hermes/plugins/ to extend functionality without forking.

Native Anthropic support : direct Claude API calls, faster and cheaper.

Voice mode : microphone input and spoken replies.

Browser direct : CDP‑based Chrome control for page scraping, form filling, and analysis.

Concurrent tool execution : parallel tool runs reduce latency from several seconds to about one second.

Why Use Your Own AI Agent?

ChatGPT excels at ad‑hoc queries, but repetitive workflows—code review, document generation, long‑term projects, and tasks requiring persistent context—benefit from an agent that remembers you. Hermes fills that gap by evolving with your usage.

Preparation

Supported systems: Linux (Ubuntu 20.04/22.04/24.04 recommended), macOS 12+ (Intel & Apple Silicon), and WSL2 for Windows users. Recommended hardware: ≥2 CPU cores, ≥4 GB RAM (8 GB+ preferred), ≥10 GB disk, and internet access to GitHub.

The only required dependency is git. Install it with:

# Check git version
git --version
# Ubuntu/Debian
sudo apt install git
# macOS (Homebrew)
brew install git

Chinese users may configure Git, pip, and npm mirrors to speed up downloads.

One‑Click Installation

Run the official script:

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

The script detects architecture, installs uv, Python 3.11, Node.js v22, clones the repository, creates a virtual environment, installs dependencies, and registers the global hermes command. Installation typically takes 3–10 minutes.

Afterward, reload your shell:

# Bash
source ~/.bashrc
# Zsh (default on macOS)
source ~/.zshrc

Verify with hermes --version (e.g., hermes v0.3.0) or run hermes doctor for a health check.

Model Configuration

Run hermes setup for an interactive wizard that asks for provider, API key, and basic preferences. Alternatively, switch models directly with hermes model.

Supported providers:

Nous Portal : OAuth login, zero‑config subscription.

Anthropic Claude : API key or authorization, native support.

OpenRouter : API key, >200 models, high flexibility.

OpenAI : API key, mature ecosystem.

Ollama : local model execution, completely free.

Recommendations: Claude or OpenAI for best performance; OpenRouter for cost‑saving; Ollama for fully local usage.

First Interaction

Start a conversation with hermes. The terminal shows a colorful UI, command history, and usage statistics. Common commands include: /new or /reset – start a new chat. /model [provider:model] – switch the LLM. /personality [name] – set a persona. /retry – retry the last request. /undo – undo the last message. /compress – compress context to save tokens. /usage – view usage statistics. /skills – list available skills. /help – show help.

Demonstration of self‑evolution: ask Hermes to write a Python script, specify a naming style, then ask it to write a similar script. Hermes remembers the preferred naming convention, comments, and structure.

Advanced Usage

Cron jobs : describe a periodic task in natural language (e.g., “daily 9 AM summary”), and Hermes generates the appropriate crontab expression.

Message‑platform integration : set up gateways for Telegram, Discord, Slack, or WeChat with hermes gateway setup and hermes gateway start. The agent can receive commands and return results on those platforms.

Browser collaboration : connect via CDP with /hermes browser connect to scrape pages, fill forms, analyze structure, or run automated tests.

Migrating from OpenClaw

Run hermes setup (auto‑detects OpenClaw) or manually trigger migration with hermes claw migrate. The process transfers persona files, memory files, skills, API keys, and gateway configurations.

Directory Structure

~/.hermes/
├── config.yaml      # main config (model, terminal, TTS)
├── .env             # API keys (do not commit)
├── auth.json        # OAuth credentials
├── SOUL.md          # agent personality
├── memories/        # persistent memory files
├── skills/          # auto‑generated skill files
├── cron/            # scheduled tasks
├── sessions/        # gateway session logs
└── logs/            # error and gateway logs

Management commands: hermes config, hermes config edit, hermes config check, hermes update.

Troubleshooting

If the install script times out, use the mirror configuration or clone manually:

git clone https://github.com/NousResearch/hermes-agent.git
cd hermes-agent
bash scripts/install.sh

Run hermes doctor for diagnostics; consult GitHub issues if problems persist.

Conclusion

Hermes Agent offers a smooth three‑step workflow: install, configure the model, and start using. Unlike agents that merely stack features, Hermes focuses on deep, self‑evolving capabilities that become indispensable for repetitive, long‑term workflows. If you have tasks that can benefit from accumulated experience, deploying Hermes is well worth the effort.

GitHub: https://github.com/NousResearch/hermes-agent

Documentation: https://hermes-doc.aigc.green