Hermes Agent
An AI that gets better the more you use it — without opening a terminal once
Hermes Agent is the open-source assistant from Nous Research that writes down how it did a job and does it better next time. The reviews all end the same way: not for ordinary users yet, because installing it means a terminal, API keys and permissions on a server. This page is those three things done for you. What is left is three steps, each with a picture.
Why not just keep using ChatGPT
Picture two new hires. The first arrives every morning remembering nothing of yesterday and has to be taught again. The second writes everything in a notebook, turns the work into a routine, and gets better the longer they stay. Most AI assistants are the first. Hermes Agent was built to be the second.
What is different from the chat you already use:
- It does work, not just answers — reads and writes files, runs code, and runs jobs on a schedule you set.
- It remembers — work it has done, what worked, what you prefer. You do not start over each time.
- It teaches itself — a complicated job is written down as a “skill” and reused.
- It is yours — you choose the model, you can change provider, and you can move the whole thing elsewhere with its memory.
So why is almost nobody using it
Because the usual install guide opens with “open a terminal on your computer”, and goes on to API keys and to limiting what an agent that can run commands may touch on your machine — which is a real thing to be careful about.
On Erawan those go away like this:
What you normally do yourself | On Erawan |
|---|---|
Open a terminal and type install commands | Press one button in a browser |
Find a server, set up HTTPS | Your own https address, straight away |
Keep the agent away from the files on your computer | It runs in its own box in the cloud — it is not on your computer at all |
Set up the program's security yourself | Hermes limits its own permissions from the moment it starts; nothing to set |
Step 1 — Get your Hermes, and set a username and password
Press this link to get your Hermes. If you have no Erawan account yet, you are taken to sign up first.
It stops to ask for two things, and you invent both. This is the name and password for getting into your Hermes — nothing to do with your Erawan account.
Field | What to enter |
|---|---|
HERMES_DASHBOARD_BASIC_AUTH_USERNAME | A username you will remember |
HERMES_DASHBOARD_BASIC_AUTH_PASSWORD | A password — make it long; let a password manager generate it |
It really does need to be long: this page is the one door to an agent that can do work on your behalf. Once both are in, the deploy carries on by itself. Wait for the status to say running, then open your address.
Step 2 — Create a key for the model (OpenRouter)
Hermes is the agent. The “brain” is a language model you choose. The simplest way in is OpenRouter: one account that works with nearly every model, paid as you use it.
Sign up or sign in, go to Keys, and press Create Key.
Name the key anything, and set a credit limit — this key is your money, and the limit is the most it can ever spend.
Copy the key straight away — OpenRouter shows it once. If you have never added credit, add a little on the Credits page first.
If you already pay for ChatGPT and would rather not open another account, skip this step — in step 3 there is a button to sign in with your ChatGPT account instead of using a key.
Step 3 — Set Hermes up until it answers
Open your address. You will see Hermes' own sign-in page; use the username and password from step 1.
Open Keys, choose the Providers tab, find OpenRouter, and paste the key from step 2. Hermes saves it to your app's storage at once — nothing to restart, and it is still there after an update.
Then open Models and choose one. The most capable model is also the most expensive — start with something in the middle; you can change it whenever you like.
Open Chat. It looks like a dark console rather than the white chat box you are used to, and it works exactly the same way: type ordinary language where it says Ask me anything… and press Enter.
Give it a real job — for example, “Remember that my shop sells coffee, opens 8 to 5 and is closed on Mondays, then draft three replies for a customer asking about opening hours.” Ask about the shop again tomorrow; it still knows.
Some tools — searching the web, generating images — need a key for that service too. They go on the same Keys page, and until then Hermes tells you itself that the tool is not available.
What it costs, and what is yours
One Hermes running all the time fits Erawan's Hobby plan — ฿99 a month, a fixed price. The model is paid to OpenRouter by use, and capped by the limit you set yourself.
Its memory, your conversations, the skills it builds and your key all live in your app's own storage. The program is entirely Nous Research's — none of our code is inside it — so the day you want to run it somewhere else, it moves whole, memory included.
Going further
- Talk to it in the chat app you already use — Hermes connects to LINE, Telegram, Discord, Slack and more from its Channels menu. We are walking the LINE steps on Erawan end to end before writing them up as their own lesson.
- An easier face — Hermes Desktop is an app for macOS, Windows and Linux with a “Connect to existing Hermes” mode for a Hermes running elsewhere. We have not tested that mode against Erawan yet; it gets its own lesson once we have.
- Your own address — swap
.erawan.appfor a domain of yours. Here is how.