One Line to Install. About an Hour to Your First Real Task.

Hermes Agent installs in one command and gets you talking to it in about ten minutes. Getting from there to something useful, a bot on Telegram or a scheduled job, takes about an hour. This is that hour, in order, with the five places people get stuck.

01. Install it

one line, then reload your shell

On macOS, Linux, WSL2, or Termux:

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

On Windows, in PowerShell:

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

There is also a desktop installer on the site if you would rather not use the terminal. The basic CLI needs no Node, Python, or Docker. Docker only comes in later if you want a sandboxed terminal, and voice mode has its own extra install step. It is a curl-to-shell install, so open the script in a browser and read it first if that matters to you.

02. Run the setup wizard

the fast path and the full path

The fast path signs you in to Nous Portal and turns on the Tool Gateway in one step:

hermes setup --portal

One Nous subscription then covers 300-plus models and the Tool Gateway, so you are not collecting API keys. If you would rather bring your own provider, run the wizard without the flag:

hermes setup

That gives you three modes. Quick is the sensible defaults. Full walks every option. Blank Slate starts with everything off except a provider, a model, and the file and terminal tools, which is the one to pick if you want to add capability deliberately rather than turn things off later.

Config lands in ~/.hermes/config.yaml. Secrets and tokens go in ~/.hermes/.env. You can set anything from the command line too:

hermes config set model anthropic/claude-opus-4.6
hermes config set terminal.backend docker
hermes config set OPENROUTER_API_KEY sk-or-...

03. Pick a model, and the one rule

64,000 tokens of context, minimum

Change provider or model any time with an interactive picker:

hermes model

The one hard rule from the docs: use a model with at least 64,000 tokens of context. Hermes injects your memory, skills, and context files into every message, and a small window starves it. Below that line you get truncated, confused replies that look like a Hermes bug but are a context problem.

Leave model routing off at the start. It is aggressive by default and will sometimes pick a model that is not actually available to you, which reads as "model unavailable" for no clear reason.

04. Start the CLI, run the first task

one real question, not "hello"

hermes           # classic interface
hermes --tui     # the modern TUI, what most people use
hermes --continue # resume your last session

Do not test it with "hi". Give it something real so you see it use a tool:

Summarize this repo in 5 bullets and tell me what the main entrypoint is.

You know it worked when the banner shows your chosen model and provider, the reply comes back with no error, and it reaches for a tool when the task needs one. Try a second one that touches your machine:

What is my disk usage? Show the top 5 largest directories.

Multi-line input is Alt+Enter, Ctrl+J, or Shift+Enter. To interrupt it mid-answer, type a new message and hit Enter, or press Ctrl+C once.

05. Wire your first chat app

one wizard covers all of them

hermes gateway setup    # Telegram, Discord, Slack, WhatsApp, Signal, Email, Teams
hermes gateway status   # check the bot is actually running

The wizard asks for a bot token for whichever platform you pick, and for an allowlist of who is allowed to talk to it. Get the allowlist right. A messaging bot with terminal access that anyone can message is a bad day waiting to happen, so never open it to everyone.

06. The slash commands you will use

learn six, ignore the rest for now

Switching model mid-conversation with /model works, but it breaks the prompt cache and the next turn re-reads the whole conversation at full input price. Pick a model per session instead of hopping.

07. When it breaks: the recovery order

run these top to bottom

hermes doctor          # diagnose
hermes model           # re-select the provider
hermes setup           # full reconfigure
hermes sessions list   # check session state
hermes --continue      # resume the last one
hermes gateway status  # check the bot

The usual causes: empty or broken replies almost always mean provider auth or the wrong model, so hermes model first. A custom endpoint returning nonsense means the base URL or model name is wrong, so test that endpoint on its own before blaming Hermes. And --continue finding no session usually means you switched profiles.

08. What to do next

one real task, then a schedule

The docs put the foundation at about an hour: install, first chat, learn the CLI, adjust config. The next two to three hours is where it gets useful: sessions, the messaging bot, tools, skills, memory, and cron scheduling.

Give Hermes one task you do every week and let it finish. It will write its first skill from that. Open the skill and read it. If it caught the right lesson, you have seen the thing that makes this agent worth running. The skill loop guide is the next one to read.