Day 1 · Install Hermes, Say Your First Word

Today’s Goal

Run hermes or hermes --tui and get a reply to your first message.

---

Prerequisites

One requirement: Git must be installed.

The installer handles everything else automatically — Python 3.11 (via uv, no sudo required), Node.js v22, ripgrep, ffmpeg. You do not need to pre-install any of these.

git --version   # any output is fine, version does not matter

If you get command not found, install Git first:

# macOS
brew install git
# or Xcode Command Line Tools (no Homebrew required)
xcode-select --install

# Ubuntu / Debian / WSL2
sudo apt update && sudo apt install git

# Fedora / RHEL
sudo dnf install git

Windows native is not supported

Hermes supports Linux, macOS, WSL2, and Android (Termux). Native Windows is not supported. Install WSL2 first, then run the commands below inside a WSL2 terminal.

---

Steps

1. One-command install

   Copy and paste this into your terminal:

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

   The installer will: download dependencies, clone the repo, create a virtual environment, and register the hermes command in your global PATH. Expect 1–3 minutes depending on network speed.

   What the installer does

   No manual pip install needed. The installer detects missing dependencies and fills them in automatically, including Python version management via uv.

2. Reload your shell

   After installation, the terminal will prompt you to run:

   source ~/.bashrc   # bash users
   # or
   source ~/.zshrc    # zsh users (macOS default)

   This makes the hermes command available in your PATH.

3. Choose a Provider

   This is the most important step. Run the interactive wizard to select an LLM provider and model:

   hermes model

   Situation	Recommended choice
   First time, want the fastest start	Nous Portal or OpenRouter
   Already have an Anthropic API key	Anthropic (Claude)
   Already have an OpenAI API key	OpenAI Codex
   Want a local model	Ollama or a custom OpenAI-compatible endpoint

   Minimum context requirement: 64K tokens

   Hermes requires a model with at least a 64,000-token context window. Most major models (Claude, GPT-4o, Gemini, Qwen, DeepSeek) meet this. For local models, set the context size explicitly (llama.cpp: --ctx-size 65536).

   You can switch anytime

   Provider selection is not permanent. Run hermes model at any time to switch — your configuration is preserved.

4. Start your first conversation

   Two interfaces are available — pick whichever you prefer:

   hermes          # classic CLI
   hermes --tui    # modern TUI (recommended — mouse support, popups, non-blocking input)

   After launch you will see Hermes’s welcome banner showing the active model and available tools. Try sending a message:

   Hello, list all the files in the current directory.

   Success looks like:

   • The banner shows the provider and model you just selected
   • Hermes replies without errors
   • It invokes the terminal tool to fulfil your request

5. Verify session persistence

   After exiting Hermes, run:

   hermes --continue   # resume the last conversation
   hermes -c           # shorthand

   If you land back in the previous conversation, session persistence is working. This matters when you start running parallel tasks later in the series.

---

Expected Output

After completing this chapter you should see something like:

$ hermes
Welcome to Hermes Agent v0.10.x
Provider: anthropic · Model: claude-opus-4.6
Tools: terminal, file, web_search ...

You: Hello
Hermes: Hello! I'm Hermes. What can I help you with?

---

Troubleshooting

hermes: command not found

Run source ~/.bashrc (or ~/.zshrc). If that does not fix it, add the path manually:

export PATH="$HOME/.local/bin:$PATH"

Append this line to your shell config file, then source it again.

Install script fails

Check whether you can reach GitHub. On a restricted network, try installing via pip instead:

python3 -m pip install hermes-agent

Hermes starts but replies are empty or show errors

Provider authentication or model selection is likely wrong. Re-run:

hermes model

Verify that the provider, model name, and API key are all correct.

Full diagnostics

Run hermes doctor — it lists every missing item with fix suggestions. When something strange is happening, this is the first command to run.