Hermes Agent

The free alternative toChatGPT Plus,Claude Pro,Cursor Pro

🟡 Intermediate⏱️ ~15 min to installYou save €23/mo

Hermes is an agent that learns. What sets it apart from a regular chatbot is its loop: it builds its own skills from what you have it do, improves them through use, searches its past conversations and gradually builds a model of who you are. It lives where you already are: Telegram, Discord, Slack, WhatsApp, Signal, or in your terminal. It runs on a €4 server just as well as on your machine, and it accepts any model: yours running locally, OpenRouter, OpenAI, your own endpoint. Where ChatGPT, Claude or Cursor sell you a monthly subscription for a model they picked, Hermes lets you plug in whatever you want behind it, including a local model that costs nothing and lets nothing leave your machine.

246.4kView on GitHub ↗

Let me show you

What if your AI stopped forgetting?

I break Hermes Agent down on video: what it is actually for, when to use it, and how to install it.

See the post on Instagram ↗210 000 views

Versus ChatGPT Plus / Claude Pro / Cursor Pro: the match

Price
Hermes Agent
€0, and nothing at all with a local model
ChatGPT Plus / Claude Pro / Cursor Pro
€17 to €23/mo, per subscription
The model
Hermes Agent
Whichever you want, switched in one command
ChatGPT Plus / Claude Pro / Cursor Pro
Theirs, and you follow their updates
Memory
Hermes Agent
It learns from you and keeps it all at home
ChatGPT Plus / Claude Pro / Cursor Pro
Whatever they agree to remember, on their servers

ChatGPT Plus / Claude Pro / Cursor Pro ↗

VS	Hermes Agent	ChatGPT Plus / Claude Pro / Cursor Pro ↗
Price	€0, and nothing at all with a local model	€17 to €23/mo, per subscription
The model	Whichever you want, switched in one command	Theirs, and you follow their updates
Memory	It learns from you and keeps it all at home	Whatever they agree to remember, on their servers
Where it lives	Terminal, Telegram, Slack, WhatsApp, Signal	Their app, and nothing else
Your documents	Your second brain, read locally	Uploaded to them

How to install it (even from zero)

Installation is a single command. The real work starts after: choosing your model, and above all deciding what the agent is allowed to do.

Paste this prompt into Claude (Code or Desktop)

It installs Hermes from the official repository and guides you to the model that fits your machine:

Installe Hermes Agent sur ma machine (l'agent IA open source de Nous Research, repo GitHub NousResearch/hermes-agent) en suivant la procédure officielle de hermes-agent.nousresearch.com. Une fois installé, aide-moi à choisir mon modèle avec la commande hermes model : dis-moi si ma machine peut faire tourner un modèle local et lequel, sinon configure un fournisseur avec ma propre clé API. Vérifie enfin que la commande hermes démarre bien et que je peux discuter avec lui. Explique chaque étape simplement, je débute.

Review the GitHub repository and check whether it is safe to install given my machine and my infrastructure. Ask me as many questions as possible, using widgets, to clarify my request and be as precise as you can.

Decide what it is allowed to do

Hermes ships with many tools enabled, including sending messages and web access. Before trusting it with anything serious, switch off what you do not need. Read my take below, there is a trap that cost me.

Dans mon installation de Hermes Agent (repo NousResearch/hermes-agent), je veux restreindre les outils disponibles. Va lire le dictionnaire TOOLSETS dans le code source du projet pour connaître la liste RÉELLE des outils, ne te fie pas à la documentation ni à la liste proposée par l'interface qui n'en est qu'un sous-ensemble. Puis configure une liste blanche : nomme uniquement les outils que je garde et coupe tout le reste. Termine en me prouvant l'état effectif avec la commande hermes tools --summary, et redis-moi le nombre d'outils actifs.

Review the GitHub repository and check whether it is safe to install given my machine and my infrastructure. Ask me as many questions as possible, using widgets, to clarify my request and be as precise as you can.

3
Wire it to your second brain
This is where it gets interesting: give it access to your notes folder. It searches them, writes to them, and its memory builds on YOUR material rather than on nothing.

What you actually do with it

An assistant that actually knows you

As it works with you, it retains your way of doing things, your projects, your habits. After a few weeks, you stop re-explaining everything in every conversation.

It answers on Telegram while you walk

The agent runs on a small server and you talk to it from your phone, by message or voice note. It works while you do something else and reports back.

Scheduled tasks in plain language

A report every morning, a backup every night, a watch every week: you ask in plain language and it runs on its own, without you.

Project health, at a glance

Before installing an open-source tool, one question matters: is the project alive? These scores are computed from public GitHub data. They are not made up.

94/100

Excellent

Very active, well-maintained project: go ahead with confidence.

Activity: 100
Maintenance: 100
Community: 100
Maturity: 70
Momentum: 100

The repo in numbers

GitHub stars: 246.4k
Forks: 51.5k
Contributors: 400
Commits: 36.4k
Project age: 1.2 yrs
Last commit: today
Size: 954.5 MB
License: MIT

Built with

Python 72.2%
TypeScript 24.7%
JavaScript 1%

Scores computed automatically from the GitHub API (last-30-days activity, update freshness, community size, project age). Refreshed daily.

Loucash’s take

10/10

This is the most obvious tool in the whole library, and the only one I give a 10. Deploying a Hermes wired to your second brain, with a local model behind it, is today the most direct way to have a personal agent that knows you and tells your life to nobody. Everything you pay for elsewhere by subscription, you get here in one command. This is not an idle recommendation: every one of my B2B clients has one. I systematically ship a Hermes in the backend of their system, it has become the default building block of my engagements. When a tool moves from testing to production at paying clients, that is the strongest signal I can give you. One trap I paid for, so you will not. Installing Hermes at a client, I had switched off the dangerous tools by copying names from the documentation. Three things escaped me at once: an unknown tool name is ignored WITHOUT any error, so the tool stays active; the name I had guessed was wrong; and the list the interface offers is only a subset of the real registry, so message sending was running enabled on a client's server. Seven active tools instead of the five I wanted. Remember two rules: a security restriction is read in the source code, never in the docs, and it is written as an allowlist. Then prove the real state with hermes tools --summary instead of trusting your config. That is why the second install step above exists.

It’s for you if: You pay for one or more AI subscriptions and want an agent that belongs to you, learns from you, and can read your second brain without sending anything out.