Hermes agent ssh earns a 7.8 out of 10 from me: it is the strongest way to let Hermes work on a real server while keeping its safety check and rollback, held back mainly by key-only logins and the fact that the server is not a sandbox.

This verdict rates the SSH terminal backend against nine criteria, each scored from what the official Hermes configuration reference and related docs actually say.

Each rating carries a one-line reason, so you can see exactly what earned or cost the points.

If you want a view on whether it fits your own setup, book a free SEO strategy session with my team.

What Was Reviewed

The subject is terminal.backend ssh, one of six built-in Hermes terminal backends alongside local, docker, modal, daytona and vercel_sandbox.

It sends each shell command the agent issues to a remote host, while Hermes itself keeps running where it was installed.

Configuration is a backend switch in config.yaml plus TERMINAL_SSH_HOST and TERMINAL_SSH_USER in ~/.hermes/.env, with optional port, key and persistence variables.

The Desktop SSH gateway connection is reviewed separately near the end, because it is a different feature that happens to use SSH too.

The review weighs the backend for the job most people buy it for: an agent doing maintenance, scripting and file work on a server that already exists.

Criteria were chosen around what goes wrong in that job: lost state, risky commands, leaked keys and confusing behaviour between the two machines.

The Scorecard

CriterionRatingWhy
Setup effort9/10One hermes config set command plus two required env lines.
Safety check kept9/10The dangerous-command check stays on, unlike every container backend.
Undo and checkpoints9/10Checkpoints and /rollback work on SSH; container backends get neither.
Secret handling8/10Values ride OpenSSH SendEnv and provider keys never leave, but a missing AcceptEnv fails silently.
Shell state between commands7/10Working folder and exports persist; aliases, functions and plain variables do not.
Isolation6/10A separate machine, but not a sandbox: changes on the server are real.
Login options6/10Key auth only with BatchMode=yes; no jump-host or multi-host option is documented.
Desktop integration8/10SSH gateway connections stay live since v0.23.0, including Windows hosts.
Documentation clarity8/10Every key is documented, though the backend and the Desktop feature share a name.
Overall7.8/10Average of the nine criteria above.

The three nines all come from the same place: SSH keeps the protections that container backends drop.

The two sixes reflect limits you plan around rather than flaws in what is there.

Strength: It Keeps The Dangerous-Command Check

The backend security comparison lists the dangerous-command check as kept for both local and ssh.

Every container option in that table, from Docker to Modal to Vercel Sandbox, skips it, on the reasoning that the container is the boundary.

For work on a live server, that difference is the single biggest reason to pick SSH.

A risky command still pauses for a person, even though it is about to run somewhere else.

Consider a maintenance job that cleans old log folders on a web server.

On Docker the cleanup would run inside a container and the check would not fire; on SSH it targets the real server and the check stays in the loop.

That is the right default for any machine customers depend on.

Strength: Rollback Still Works

The checkpoints page is clear that container backends record no checkpoints and that /rollback refuses diffs and restores there.

SSH sits with local on the side that keeps them.

That means a config file the agent rewrites on the server can be put back from the Hermes side.

Few remote-execution setups give you an undo path this direct.

It does not replace server backups, since checkpoints cover the agent's file writes rather than the whole machine.

Still, for the common mistake of a bad edit to a config file, it turns a repair job into one command.

Strength: Credentials Stay Where They Belong

Skill variables reach the server through OpenSSH SendEnv, so their values never appear in the command text.

Provider keys such as OPENAI_API_KEY are never forwarded, even when listed for passthrough.

The tools reference also routes SUDO_PASSWORD to whichever backend runs the command, which on SSH means the remote host.

The design keeps model access on the Hermes machine and only sends the server what a job needs.

That is why secret handling scores an eight and not lower.

๐Ÿ”ฅ Want to see this SSH setup running inside a full agent stack before you build your own? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS with Hermes, four coaching calls a week and 3,600+ members comparing backends and server layouts. โ†’ Join the AI Profit Boardroom

Strength: The Connection Model Is Efficient

Hermes holds one SSH ControlMaster connection open and reuses it, keeping it alive for five idle minutes.

It captures the remote login environment once, then every command sources that snapshot and moves into the tracked folder before running.

The result is a consistent PATH on each call without the cost of a full login every time.

Host keys are handled with StrictHostKeyChecking=accept-new, so a new host is trusted and stored on first contact.

Limit: Shell State Is Partial

The persistent shell section describes what carries over with terminal.persistent_shell on, its default.

The working directory and exported variables persist.

Aliases, functions and plain variables set mid-session do not.

Agents that define a helper alias in one step and call it in the next will trip over this, which is why the criterion lands at seven.

The fix is simple once you know it: export values you need later, or put helpers in the remote login profile so the startup snapshot picks them up.

Turning persistence off entirely is possible with hermes config set terminal.persistent_shell false, though few setups will want that.

Precedence is documented clearly: TERMINAL_SSH_PERSISTENT first, then the config key.

Limit: Key-Only Logins And One Host

Connections open with BatchMode=yes, so password logins are not part of the documented path.

The docs list a single host, user, port and key, and describe no jump-host, agent-forwarding or multi-host option.

Teams with bastion setups will need their own SSH configuration outside what Hermes documents.

That scope is reasonable, but it caps the login options score at six.

Port and key are both optional, with port 22 and the system default key used when they are left out.

Anyone managing several servers will be running one setup per host or using the Desktop gateway route instead.

Limit: It Is Not A Sandbox

Isolation scores six because the server is a real machine the agent can change.

The tools page calls SSH recommended for security because the agent can't modify its own code, which is a different promise from containment.

For containment, Docker or Modal are the tools, and they cost you the check and the checkpoints.

The security docs suggest running the gateway on a separate machine or VM for maximum isolation.

In practice, the protection you get comes from the login user: give it access to the folders and services the agent needs, and nothing else.

That one habit moves the isolation picture further than any Hermes setting can.

A Quirk Worth Knowing: The Bang Shortcut

CLI commands typed with a leading ! always run on the Hermes host, according to the CLI docs.

So !hostname names your laptop while the agent's terminal names the server.

Path completion in the composer, by contrast, follows the SSH backend.

It is documented and logical, but it confuses new users, and it is part of why documentation clarity sits at eight.

๐Ÿ“ž Weighing SSH against Docker or Modal for your own agents? Book a free SEO strategy session and my team will help you pick the backend for each job. โ†’ Get a free second opinion

The Desktop SSH Connection, Rated On Its Own

Hermes Desktop can add an SSH connection under Settings โ†’ Gateways, entered as user@host:22.

It drives a full remote Hermes install: the app opens a tunnel, starts the dashboard and adopts a token, per the multi-connection docs.

The v0.23.0 release made every registered gateway stay connected, so switching is instant, and added Windows SSH hosts.

A Desktop-owned remote backend retires after 900 idle seconds by default, set by dashboard.ssh_isolated_idle_grace_s.

After an update, the update guide says to reconnect Desktop for SSH backends.

Taken alone, I'd rate it an eight: strong for multi-machine work, with one extra step after every update.

The multi-gateway connections guide covers that side in full.

Who Should Use It

  • Anyone whose real work lives on a VPS, home lab or build server.
  • Teams that want approvals and rollback kept while the agent works remotely.
  • Operators who already log in to their servers with keys.

A typical fit is a freelancer running client sites on one VPS, who wants the agent to read logs, apply small config edits and report back.

Another is a developer with a home lab box who would rather the agent never touched the laptop that holds Hermes.

In both cases the work is already on the server, so the SSH backend simply meets it there.

Who Shouldn't

  • Anyone who wants a throwaway, contained environment: Docker or Modal fits better.
  • Setups that require password logins or jump hosts managed by Hermes itself.
  • People who only need help on their own desktop, where local is simpler.

Inside the AI Profit Boardroom, members compare these choices on the weekly coaching calls.

My Verdict On Hermes Agent SSH

At 7.8 out of 10, the SSH backend is the best-rounded option in Hermes for server work.

It keeps the safety check, keeps rollback and keeps model keys at home, which no container backend matches.

It asks you to accept key-only logins, partial shell state and full responsibility for the server.

For most operators with a server, those are fair terms.

The score would rise with documented jump-host support and fuller shell state, and nothing else on the scorecard needs fixing.

If you want a second view on your layout, book a free SEO strategy session and bring your server list.

Also On Our Network

FAQ: Hermes Agent SSH Verdict

What score does Hermes Agent SSH get?

7.8 out of 10, the average of nine criteria rated from the official docs. Isolation and login options are its weakest areas at 6 each.

What is the biggest reason to choose the SSH backend?

It keeps the dangerous-command check and checkpoints while commands run on a remote server; the container backends drop both.

Does the SSH backend support password logins?

The documented path is key-based. Hermes connects with BatchMode=yes, using TERMINAL_SSH_KEY or your system default key.

Is the Hermes SSH backend a sandbox?

No. It runs commands on a real machine you manage. For a contained environment, the docs offer Docker, Modal and other container backends.

How does the Desktop SSH connection rate?

About 8 out of 10 taken alone. It keeps remote installs connected since v0.23.0, but needs a Desktop reconnect after each update.

Who should skip Hermes Agent SSH?

People who want a throwaway container, need Hermes-managed jump hosts or password logins, or only work on their own computer.

๐Ÿ“บ Video notes + links to the tools ๐Ÿ‘‰

๐ŸŽฅ Learn how I make these videos ๐Ÿ‘‰

๐Ÿ†“ Get a FREE AI Course + Community + 1,000 AI Agents ๐Ÿ‘‰

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

  • I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
  • I've grown a YouTube channel to 400,000+ subscribers.
  • I wrote "Link Building Mastery".
  • My Udemy courses have taught over 29,000 students.

โ†’ Get my best AI training inside the AI Profit Boardroom

Set up key login, switch the backend, test one reversible job and judge it on a week of real work, and if you want help, book a free SEO strategy session or join the AI Profit Boardroom to compare notes on hermes agent ssh.