What Is Hermes Agent? How to Install (Mac, Linux, Windows)

Hermes Agent is an open-source AI agent developed by Nous Research. It is published on GitHub under the MIT license and runs in the terminal, in a desktop app, in messaging apps and in editors. What sets it apart from an ordinary chatbot is that it learns from the work it does, writes reusable skills, and keeps its memory across sessions.

This post is the first part of a series: I explain what Hermes is, how to install it and what to configure on first run. In the second part I cover Telegram, scheduled tasks and safe use.

This post is based on the official documentation and the GitHub repository; I have not tried the installation on my own machine. I pass on the commands below as they are, with sources at the end. I deliberately leave out version numbers because they vary by source.

What is Hermes Agent?

Hermes is an agent framework built around a language model. It is not a model; you choose the model. The framework provides:

• Persistent memory. Sessions are kept in a SQLite database (with full-text search), so a conversation from weeks ago can be recalled. Your preferences are stored in files such as memory.md and user.md.
• A skill system. After finishing a complex job, the agent can extract a reusable procedure from it and improve that procedure as it uses it. Skills live under ~/.hermes/skills/ and are compatible with the agentskills.io format.
• A multi-platform gateway. You can talk to the same agent over Telegram, Discord, Slack, WhatsApp and Signal; you can continue a session you started in the terminal from Telegram.
• Subagents. It can open parallel workflows with the delegate_task tool.
• MCP support. It can connect to Model Context Protocol servers.
• Different execution environments. Local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox.

The repository also includes the hermes claw migrate command for migrating from OpenClaw; it carries over personalities, memory, skills and settings.

I explained the skill concept in detail earlier in What Is an Agent Skill? How to Write Useful Skills. Hermes is one of the tools that applies this idea most directly.

Requirements

According to the official installation page, the only hard requirement is Git. The installer script handles Python, Node.js and the other needed tools itself. Supported environments:

Environment	Status
macOS	Desktop package is Apple Silicon only; command-line install also works on Intel
Linux	Command-line install
Windows	PowerShell installer or WSL2
Android	Termux package (aarch64)

If you will run a local model on your own hardware, you also need GPU memory; if you use a cloud model you do not. A third-party guide notes that the agent expects at least 64K tokens of context from the model; if you pick a model with a small context, you may get errors at startup.

Installation

macOS, Linux and WSL2

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
source ~/.zshrc   # ~/.bashrc if you use bash
hermes

The alternative address in the GitHub repository serves the same script: https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh.

Windows

In PowerShell:

iex (irm https://hermes-agent.nousresearch.com/install.ps1)

Alternatively, you can install WSL2 and run the Linux command in an Ubuntu terminal. Third-party guides describe native Windows installation as an early stage (beta); if you run into problems, WSL2 is the more solid route.

Desktop app

There is also a downloadable desktop package for macOS and Windows; the macOS version supports only Apple Silicon. You can also start from the command line and open the local app with the hermes desktop command.

Check the installation

hermes doctor

This command checks system health. If you get a hermes: command not found error, restart the terminal or source your shell file; the installer puts the command-line entry at ~/.local/bin/hermes. Do not install with sudo.

First setup

hermes setup

The setup wizard asks for your model provider. Options include Nous Portal, OpenRouter, Anthropic, OpenAI, Google Gemini and local models such as Ollama. To change it later:

hermes model       # choose provider and model
hermes tools       # turn tools on/off
hermes config set FIRECRAWL_API_KEY your_key

Two rules matter:

1. Secrets live in ~/.hermes/.env, settings live in config.yaml. The documentation explicitly says not to put a setting in the .env file.
2. Change settings with hermes config set; do not edit the YAML by hand.

Files are gathered under ~/.hermes/: things like skills/, state.db (the session store), sessions/ and logs/.

Which model should you run it with?

Hermes itself has no fee; the cost comes from the model you choose.

• Cloud API (OpenRouter, Anthropic, OpenAI, Gemini): Easy to set up, pay as you go.
• Nous Portal: Nous's own subscription; you sign in with hermes setup --portal.
• Local model with Ollama: Free and offline. Example flow: download a model, run ollama serve, then in hermes model pick "Custom endpoint" and enter http://localhost:11434/v1. Because Ollama's default context window is small, you may need to enlarge it with the OLLAMA_CONTEXT_LENGTH variable.

On how to use newly released models in Hermes, I mentioned in, for example, the LongCat 2.5 Preview install guide that there is a ready-made instruction for Hermes Agent. For models not yet open to access, such as Gemini 4 Argon, the model ID has to be published first.

First run

hermes

A chat opens in the terminal. Start with something small, like "what's in the folder I'm in?" Helper commands you will use:

hermes chat -q "one-off question"   # single question without opening a chat
hermes --continue                    # continue the last session
hermes --tui                         # terminal interface
hermes update                        # update

If you want to set up multiple personalities for a single workflow, there are also profiles such as hermes profile create work --clone; each profile uses its own ~/.hermes/profiles/<name>/ folder.

Who is it for?

• People comfortable in the terminal. The tool is terminal-centric; there is a desktop app too, but the real power is on the command line.
• People who also want to use the same agent from their phone. The Telegram gateway is for that.
• People who want to automate repetitive work. Skills and scheduled tasks are for that.
• People who want to stay in control. It is open source and the model choice is yours.

It may be heavy for a simple one-off code question; a direct chat tool is enough for that.

Frequently Asked Questions

Is Hermes Agent free?

The software itself is MIT-licensed and open source. What you pay for is the API fee of the model you use; if you use a local model with Ollama, it can run for free.

Does Hermes Agent work on Windows?

Yes. The official documentation provides the PowerShell installer; you can also use the Linux command inside WSL2.

What is the difference between Hermes Agent and Claude Code?

Hermes is a model-independent agent framework with memory and skill generation; it also works from messaging apps. Claude Code is Anthropic's own coding agent. For a tool comparison, see the vibe coding tools comparison.

Which models does Hermes support?

Nous Portal, OpenRouter, Anthropic, OpenAI, Google Gemini and any custom OpenAI-compatible address, including Ollama.

How long does installation take?

Third-party guides say 2 to 5 minutes; it varies with your connection speed.

Sources

• Nous Research, hermes-agent (GitHub): features, install commands, MIT license, OpenClaw migration.
• Hermes Agent, Installation: requirements, install locations, troubleshooting.
• Hermes Agent, official documentation: ~/.hermes structure, configuration rules, providers.
• DataCamp, Nous Research Hermes Agent: Setup and Tutorial Guide: memory and skill architecture, Ollama setup, profiles.
• Hermes Atlas, How to Install Hermes Agent: operating system notes and troubleshooting (third party).
• NxCode, Hermes Agent Tutorial: context window requirement (third party).