Hermes Agent: installing and configuring

By Ivo Donker · October 28, 2026 — created with AI assistance (writer: DeepSeek V4.1 Flash; critic: Gemini 3.8 Flash)

Hermes Agent: Review News

Status September 2026, compiled from our own notes and news archive; text machine-generated and editorially checked. Check versions and commands in the official documentation.

In this guide you install Hermes Agent, the open-source agent from Nous Research that remembers, learns skills by itself and is also reachable via chat apps. You choose a model, store your keys securely and check that everything works. Everything comes from the official documentation (status September 2026).

What you need

Operating system: Linux, macOS on Apple Silicon, Windows, WSL 2 or Android (Termux). Macs with an Intel processor are not supported.
Access to a language model: an API key from a provider or an OpenAI-compatible endpoint, for example a local model or your own router.

You do not need to install Python, Git and the uv package manager yourself: the installer takes care of that, with Python 3.14.

Installing

Linux, macOS and WSL 2:

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

Windows (PowerShell):

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

Then reload your shell so that the hermes command is found:

source ~/.bashrc      # of ~/.zshrc op een Mac

Configuring with the wizard

The easiest way to configure everything:

hermes setup

The wizard goes through the model, provider and the main options. Choosing or switching only the model can also be done separately:

hermes model

Where everything is stored

Hermes stores everything in ~/.hermes/ (on Windows in %LOCALAPPDATA%\hermes\; with HERMES_HOME you choose a different location):

config.yaml: model, terminal backend and other settings.
.env: API keys and other secrets.
SOUL.md: the identity and tone of your agent.
memories/: what the agent remembers about you and your work.
skills/: skills that the agent creates itself or that you install.

Rule of thumb: secrets belong in .env, everything else in config.yaml. The hermes config set command handles this automatically: names in UPPERCASE_WITH_UNDERSCORES go to .env.

hermes config set ANTHROPIC_API_KEY sk-ant-...
hermes config set OPENAI_API_KEY sk-...

Initial safety settings

Hermes can execute shell commands and by default runs on your own machine. For a cautious start, put this in ~/.hermes/config.yaml:

agent:
  dangerous_command_approval: true   # vraag toestemming voor risicovolle commando's
skills:
  write_approval: true               # laat je eerst zien wat de agent aan skills wijzigt
  guard_agent_created: true          # scan zelfgemaakte skills op gevaarlijke patronen

Do you want Hermes to run commands in a sandboxed environment instead of directly on your system? See the tutorial about the Docker backend.

Checking that it works

hermes doctor

This checks your installation, configuration and keys. Then start a conversation:

hermes

Give a simple instruction, such as "Summarise the README in this folder", and see whether the answer is correct.

Optional: Hermes via Telegram or Discord

Hermes can also receive messages via more than twenty chat platforms. The wizard:

hermes gateway setup

By default nobody may talk to your bot, except users you explicitly allow or approve via a pairing code. Leave it that way: a bot with access to your terminal should not be open to everyone.

Updating and uninstalling

hermes update --check   # kijken wat er nieuw is
hermes update           # bijwerken; draaiende gateways worden herstart

When updating, Hermes first creates a snapshot and rolls it back if the new code is incorrect. You can uninstall with hermes uninstall. Your configuration and data are retained, unless you pass --full.