Hermes Agent hosting: how to run Hermes 24/7

To run Hermes Agent 24/7, keep its messaging gateway on a machine that stays awake, supervise the process, preserve its data, and test the complete route from an incoming message to a reply. A VPS or managed host can remove your laptop from that route. It does not remove model outages, expired credentials, or the need to check recovery.

This guide uses the official Hermes documentation checked on September 5, 2026. The commands describe a Linux installation with systemd. They are a deployment recipe to verify on your host, not a report of a benchmark or a tested uptime guarantee.

What needs to keep running?

Hermes's messaging gateway receives messages and runs scheduled jobs. The interactive terminal session is a separate way to use the agent. For an unattended assistant, configure and supervise the gateway. Closing an interactive chat should not be your deployment procedure. Hermes messaging gateway documentation

Think through a concrete request: a message arrives, the agent calls a model, a tool reads a file, and the reply reaches the same chat. Every dependency in that route needs to remain available. A green server status says little about a revoked bot token or a file that exists only on your laptop.

Keep model hosting separate from agent hosting, too. If Hermes calls an external model API, the server runs the agent and its tools. You do not need to place the model's weights on that server. If you choose a local model endpoint, include that machine in your availability plan.

Choose the host around the tools you need

Host	A reasonable fit	What you need to operate
A computer you already own	Personal workflows that depend on local files or desktop applications	Power, sleep settings, network access, storage, and restarts
A Linux VPS	An assistant whose tools work on Linux and call reachable services	OS maintenance, process supervision, data protection, and access controls
Managed agent hosting	A workload compatible with the provider's supported runtime and tools	Agent permissions, credentials, budgets, and the provider's documented limits

Start with the work, then choose the machine. A browser task may need more memory than a text-only assistant. A local desktop integration may require a different operating system or an explicitly supported remote connection. Test the busiest expected workflow before buying a larger plan or committing to a long contract.

For a managed service, ask which runtime version it runs, whether you can export your data, and what happens after a failed update. A provider's list of supported agents does not establish compatibility with every plugin, local path, or authentication method.

The local, VPS, and managed hosting comparison helps assign maintenance and recovery responsibilities before choosing a provider.

Try managed Hermes hosting

If your workload fits Hermes and you want a managed host, cloudrun's project setup provisions a Hermes runtime and opens it on the canvas. Start by sending a test message, connect the required service, and verify one representative tool action before enabling scheduled work.

Check the runtime and tool requirements before moving an existing setup. You remain responsible for the agent's permissions, provider accounts and recovery plan. If you build client automations, the automation-builder setup guide explains how to scope that first workload.

Create a cloudrun account, or review the setup steps.

Install Hermes under the account that will run it

Use a dedicated account for the agent and give it access to its own workspace. Keep unrelated personal documents and infrastructure credentials elsewhere. This makes it easier to understand what a tool can reach and which files must survive a replacement host.

Follow the current Hermes installation instructions. On Linux, the documented command is:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Read the installer before executing downloaded code if your deployment policy requires review. The documentation lists Git, curl, and xz-utils as prerequisites for the Linux CLI path. After installation, open a fresh shell and configure the model and enabled tools:

hermes model
hermes tools
hermes doctor

Before enabling a messaging channel, make one small model request and one harmless tool request. If either fails, fix it here. Adding a background service will not repair the model credentials or install a missing tool dependency.

Configure the channel and background service

For a systemd user service, run these commands as the agent's account:

hermes gateway setup
hermes gateway install
hermes gateway start
hermes gateway status

An administrator must enable lingering for that account if the user service should survive logout and start at boot. Use loginctl enable-linger with the actual service account name. Hermes also documents a system-service alternative; choose one arrangement and avoid leaving both active. Hermes service management

Configure the intended user or room allowlist during channel setup. Then check two cases: an approved account can reach the agent, and an unapproved account cannot trigger its tools. Test group behavior separately from direct messages, because the audience and routing can differ.

Disconnect your SSH session and send another message. Next, reboot the host during a maintenance window and repeat the request. Check that exactly one gateway is running and that the reply comes from the expected agent.

Decide where shell commands execute

The machine that hosts Hermes and the environment that runs its terminal tools can differ. Hermes documents local, Docker, SSH, and several cloud execution backends. Its local backend runs commands directly on the host. Changing the backend changes where files, dependencies, and credentials need to be available. Hermes terminal backend configuration

Make a short inventory for each enabled tool: its executable, input directory, output directory, required credentials, and external services. Run it under the service account rather than assuming an interactive shell's environment will carry over.

If you use a sandbox, check its mounts and forwarded credentials. A container with access to a sensitive host directory can still modify that directory if the mount permits writes. Test the allowed operation and the operation you expect the boundary to reject.

If you prefer Docker, preserve the data separately

The official Hermes Docker guide distinguishes running the whole agent in a container from using Docker only for terminal tools. For the whole-agent deployment, it places persistent data in a host directory mounted at /opt/data and documents a container restart policy.

Record the image version you deploy, the volume location, and any extra packages your tools require. Recreating a container should not depend on remembering commands typed inside the previous container. Put required dependencies in a reproducible image or installation procedure.

Leave inbound ports closed unless the integration needs them. If you need a dashboard or API, follow the authentication and networking instructions for that interface. A chat connection and an administrative web interface have different access requirements.

Check useful work, not just process status

Use a small acceptance checklist before relying on the agent:

An approved user receives a reply through the actual channel.
A representative tool completes with the expected permissions.
A scheduled task runs in the intended timezone and delivers its result.
Logout and reboot do not leave the agent offline.
A failed model request appears in logs without exposing credentials.
An operator can pause the agent and identify the cause of failure.

Keep the probe inexpensive and read-only. Track the last successful result separately from host uptime. When a probe fails, inspect the channel, model, tool, and delivery stages before restarting everything.

Back up before you depend on the setup

Hermes documents hermes backup for a full archive and hermes import for restoring it. The full backup includes credentials; a profile export excludes them. Store the archive accordingly and verify the restoration procedure on a separate destination. Hermes backup and migration FAQ

Choose a backup interval based on how much new memory, configuration, and work you could afford to lose. Keep a copy outside the agent's host. A backup on the same disk helps with an accidental edit but may disappear with the machine.

When moving an existing agent, follow the migration checklist for a rehearsal and a cutover with only one active gateway.

Schedule updates when someone can inspect the result. Record the previous version, create a restore point, apply the update, and repeat the channel and tool checks. If a deployment cannot pass those checks, keep it out of unattended service until you know why.

Does 24/7 hosting guarantee instant responses?

No. A continuously running gateway can still wait for a model, hit a rate limit, or lose access to a tool. Set expectations for the complete task, including external services. For anything with a deadline, define when a person should take over and how they will learn that the agent needs help.

Budget for scheduled work separately from idle hosting. The always-on agent cost guide includes a worked example of model calls made by scheduled checks.