Hermes Desktop as a Shared Agent Workspace

Hermes Desktop is a native interface to the same agent used by the CLI, TUI, and messaging gateway. Local profiles, provider settings, sessions, skills, memory, and tool configuration remain shared rather than being copied into a separate product.

Official documentation: https://hermes-agent.nousresearch.com/docs/user-guide/desktop

Local and remote backends

By default the app launches and manages a local backend. It can also attach to a remote hermes serve or documented dashboard backend on a trusted machine. Remote mode requires a reachable backend and a real authentication provider; the messaging gateway is a separate process and must still run if Telegram, Discord, or Slack access is required.

Use the built-in file browser to follow project edits, the model picker to change configured routes, and the session/profile surfaces to keep work separated. Microphone access should be granted only when voice is intentionally used.

Desktop checklist

  • Confirm the active profile before sending a task.
  • Set the intended project directory.
  • Verify provider and tool status in settings.
  • Keep remote backends behind strong authentication and a private network.
  • Distinguish the desktop backend from the messaging gateway.
  • Review microphone and filesystem permissions.
  • Use saved sessions instead of duplicating conversations manually.
  • Check the desktop log first when boot fails.

Pitfalls

  • Assuming “backend ready” proves authenticated chat WebSockets work.
  • Exposing a remote management backend directly to the public internet.
  • Switching profiles without noticing that the working context changed.
  • Treating a gateway restart as a desktop-backend restart.
  • Removing the full Hermes installation when only the GUI should be uninstalled.

Verification steps

  1. Launch Hermes Desktop and confirm the expected profile, model, and project path.
  2. Resume a known CLI session.
  3. Make one harmless file read and confirm the browser points at the intended workspace.
  4. If remote, verify the backend reports authentication required before signing in.
  5. Test a chat WebSocket, not only the status endpoint.
  6. Check hermes logs gui -f during one controlled retry.
  7. Confirm messaging channels separately through the gateway status and a test message.