If you've been following the AI agent space, you probably noticed that things are accelerating again. Recently, Nous Research released Hermes Agent, and in some very important ways, it is better than OpenClaw.

The thing that immediately stands out is that it keeps getting more capable the longer you run it. It has persistent memory and auto-generated skills.

Each skill it develops is treated like a scientific project—forming hypotheses, testing them out, and improving its own abilities.

It learns your projects and never forgets how to solve a problem.

And the best part? It's built by Nous Research, an open-source, distributed training lab that believes big AI labs have no business setting the morality and ethics of your models.

The user should have the power to decide how the model behaves.

Yesterday, Nous Research dropped version 0.9.0—the "Everywhere Release".

If you missed the OpenClaw train, this is your chance to jump in.

Right out of the box, Hermes Agent comes with 74 installed skills.

It can research academic papers from arXiv, generate video and audio, and interact with Claude Code natively.

It even includes a god mode skill to bypass safety filters using "parseltongue techniques," and interact with everything from Minecraft to PyTorch.

It supports hundreds of models through OpenRouter, including premium options like Anthropic's Claude Opus 4.6, or free, high-performing alternatives like RCAI Trinity and Nvidia's Nemotron.

The long-term vision is staggering: truly scalable asynchronous LLM reinforcement learning, completely open-source.

You don't need to be a developer to get this running within an hour.

While you can install it on a local machine like a Mac Mini, the most reliable way to run an always-on agent is on a Virtual Private Server (VPS).

I recommend Hostinger's KVM2 service.

Two vCPU cores, 8 gigs of RAM, and 100 gigs of NVMe disk space are the sweet spot for preventing freezes while giving your agent room to grow.

Hostinger actually created a persistent Docker container specifically for Hermes Agent.

This means all of your agent's learned skills, memories, and session histories survive any container restarts or upgrades.

Once you deploy the server, you only need four things to configure it:

Once you plug those into the setup wizard, you use the browser terminal to connect.

You navigate to the Docker directory (cd /docker/hermes_agent),

spin up the interactive terminal (docker compose exec -it hermes-agent /bin/bash),

and activate the virtual environment (source /opt/venv/bin/activate).

From there, simply typing hermes wakes up your agent.

If you want to control Hermes remotely, the Telegram integration is the way to go.

You connect your newly created bot to the server using a pairing code.

But here's where it gets interesting. As the agent executes commands, especially dangerous ones, you'll be greeted by its built-in security bot asking for approval.

They named the security bot "Turth.".

For the nerds out there, that’s Elvish for "guard".

You have complete control to allow actions once, for the session, always, or deny them entirely.

And if something breaks, the .env file containing all your credentials is fully accessible using a simple nano text editor command.

Hermes is bleeding-edge tech, and wrangling the terminal is part of the job.

But with an agent this capable, it is entirely worth the effort.

Here’s my video about it.