Hermes Agent hosting: where to run it and what it needs

Hermes Agent keeps a memory that grows across conversations and writes reusable skills for itself. A server that runs it therefore needs room for that data to grow, and a backup that keeps it. Its messaging gateway is a single background process that its own installer can set up as a service, and it refuses strangers by default.

Several hosts we cover install Hermes Agent for you, and two of them also run it for you on a managed plan. This guide follows Hermes Agent's own documentation and each host's page, read on 27 October 2026.

What Hermes Agent needs from a server

Hermes Agent is an open-source agent from Nous Research, released under the MIT licence. Its documentation gives these points that matter for hosting, read on 27 October 2026:

• it installs with one command on Linux, macOS, WSL2 and Windows, and current installs run on Python 3.14
• its data, including the memory and skills it builds up, lives in one folder, ~/.hermes on Linux, which is what you back up
• the messaging gateway is a single background process; hermes gateway install sets it up as a service
• by default the gateway turns away anyone who is not on an allowlist or paired by direct message
• the install guide lists no memory or disk figure

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes gateway install            # a user service
sudo loginctl enable-linger $USER  # keep it running after you log out

For a service that starts when the server boots, the docs give sudo hermes gateway install --system instead. In Docker, they say to run the gateway as the container's main process with a restart policy, with the data folder mounted at /opt/data.

Which route fits you

Our guide to where to host an AI agent explains the routes in general. For Hermes Agent, these are the options among the hosts we cover:

• managed: Cloudways runs Hermes Agent for you, with updates it tests before release, and Hostinger's AI automation apps plan can run it too. Neither plan is in our price records yet
• one-click: Hostinger, DigitalOcean, Kamatera and Contabo install it when you order a server
• your own install: any server with 2 to 4 GB of memory, using the commands above or Docker

Hermes Agent's own install guide says current first-party installs run on Python 3.14. DigitalOcean's listing names Python 3.12.3, so run Hermes Agent's own updater after you order.

How much memory it needs

Hermes Agent's install guide gives no memory or disk figure. Contabo sells its Hermes plans from 8 GB. The data folder grows as the agent stores memories and writes skills, so leave disk space as well.

Running your own small model next to it

Hermes Agent works with your own model provider or with Nous Portal, Nous Research's service with many hosted models. It can also use a model running on the same server, but then the server needs the memory of that model as well. Our guide to running Ollama on a server without a GPU shows what small models need.

Security basics

Keep the gateway's default: only allowlisted or paired users. Its documentation does not recommend allowing all users for a bot with terminal access. It also ignores e-mail senders unless you turn e-mail pairing on.

Most of our guide to securing OpenClaw applies here too. Give the agent its own accounts and keys, set a spending limit, and restore a backup at least once.

What it costs each month

The server is a fixed monthly price and the model is paid per token. Hermes Agent itself is free under the MIT licence; model providers and Nous Portal have their own prices, by its home page. Our guide to what OpenClaw or Hermes Agent costs each month adds the two together with a calculator.

How we checked

We read Hermes Agent's documentation, its GitHub releases and each host's Hermes page on 27 October 2026. Our own measurements will be added here, with their dates, when our test bench runs them.

Back to AI agent hosting