What is Hermes Desktop? The app, the agent and remote connections

Everything below is as Nous Research’s own pages describe it on October 9, 2026.

Desktop is the interface, Hermes is the agent

The Desktop page calls it “The Hermes Agent app for macOS, Windows and Linux,” and the documentation is precise about what that means: “a native app built around the same agent you get from the CLI and the gateway — same config, same API keys, same sessions, same skills, same memory.” Under the window it starts its own local Hermes server, and a session begun in the terminal or the browser dashboard can be picked up in the app. For the agent itself, its memory, skills and models, see what Hermes Agent is.

The distinction matters when deciding where work runs. A window on your laptop can be the interface to a local agent or to a remote one, and its location does not tell you which computer holds the agent’s files or runs its tools.

Price, platforms and versions

“Hermes Desktop is the open-source Hermes Agent under an MIT license, free to download and use,” and “No account is needed to run the base app.” Signing in to Nous Portal is optional and adds always-on cloud agents, Nous’s models and tool credits. The downloads for macOS (12 or later) and Windows are installers that fetch Hermes and build the app; the macOS one is for Apple Silicon. On Linux there is no installer package: you install Hermes from the terminal and run hermes desktop, which also adds it to your application menu. The same command opens the app from any existing terminal install.

The app first shipped in Hermes v0.16.0 on June 5, 2026. Since then it has gained coding projects, a Hermes Cloud connection (July), artifacts and an SSH connection (August), Bot Mode built in and on by default (v0.21.0, August 31), a Simple and an Advanced interface and Bot Screen (late September). Nous publishes no separate version number for the app; its release notes say the app moves with the next bundled release.

What the app adds

Beyond the chat itself, the Desktop documentation lists tabs, panes and several windows; a preview rail, a file browser, an embedded terminal and a gallery of what the agent made; a Git pane that reviews, commits, pushes and opens a pull request, with a separate worktree per agent; a memory graph that shows and edits what the agent has learned; a global hotkey for quick questions and a floating always-on-top bar; voice; and settings screens for providers, MCP servers and chat apps, including a Telegram setup that creates the bot from a QR code. Plugins can add tools, themes and integrations, and the docs warn that a desktop plugin “is not sandboxed.” Locally, the agent can also use your computer in the background, clicking and typing without moving your cursor, on macOS, Windows and Linux.

Bot Mode is the team-shaped part: named agent profiles with their own roles, which talk in group rooms. It ships with the app and is on by default; it is a feature within Desktop, not another name for the whole application.

Can it connect to an agent on a remote server?

Yes, three ways, all kept in the app’s list of registered gateways:

• A remote gateway: a Hermes reachable at a web address, on your home network, over Tailscale or on the internet, signed in to with a session token or OAuth.
• SSH: a Hermes install on a machine you can SSH into; the app opens the tunnel and starts the dashboard for you.
• Hermes Cloud: an instance Nous hosts for you, found through your Portal sign-in.

The conditions are the ones a server always has. For a remote gateway, the docs say the server process (hermes serve) must already be running on that machine, and “The desktop app does not start it for you”; if you rely on Telegram, Discord or Slack, the messaging gateway must be running there too. Username and password sign-in is for a trusted network only, and the docs advise OAuth for anything reachable beyond your own machine. That advice has a history: Nous’s docs record a June 2026 campaign in which scanners found Hermes dashboards open to the internet and used them to plant an SSH backdoor, after which sign-in became mandatory on any non-local address. The tokens the app saves are encrypted with your operating system’s keyring.

Seeing and taking over the server’s screen

A chat connection to a remote agent is not control of its computer, but Hermes now offers a version of that too. Bot Screen gives each bot on a headless Linux host its own desktop, which its browser and computer use act on, streamed live into Hermes Desktop: you watch what the bot does, take over when it reaches “a login, 2FA prompt, CAPTCHA or payment step,” and hand control back. It is off by default and needs TigerVNC and Xfce on the server (the app offers to install them). Nous’s measured figures put a small server at 4 GB to run the screen and 8 GB for a comfortable takeover with a browser open, and its docs add the caveat that matters: “Screens are work surfaces, not security boundaries: the bots share the host’s user account, files and network.”

Two things still don’t follow from a remote connection. Your laptop’s own website logins stay on your laptop: whatever you sign in to on the bot’s screen is what the bot uses afterwards. And with a remote backend, computer use “acts on the same machine the bot’s screen lives on, never on the machine running Hermes Desktop.” If your question is whether a server-based agent can reuse the browser accounts on your own machine, start with Hermes browsing on a server.