Hermes Agent on a Cloud Mac: Install, Telegram Gateway and Local Models

Hermes Agent, from Nous Research, is the other open-source always-on agent people compare with OpenClaw: MIT licensed, reaches you on Telegram, Discord, Slack or WhatsApp, keeps memory between sessions, and writes itself a reusable skill when it solves something hard. Like any persistent agent it wants a machine that never sleeps and that you would not mind being wiped. This guide puts it on a dedicated cloud Mac in about twenty minutes.

Why a Mac, and why a rented one

Hermes runs on Linux, so if the agent only calls cloud models a Linux VPS is the cheaper host. A Mac earns its place when you want local inference on Apple silicon, when the agent should drive macOS tools (Shortcuts, Safari, Xcode, the iOS Simulator), or when you want the agent off the computer that holds your passwords. A rented Mac Mini M4 or M6 adds a static IP, a wired gigabit line, no hardware to buy, and a reset button. The Mac mini vs VPS post covers the trade-off in detail; it applies to Hermes just as much as to OpenClaw.

What you need

• A Mac with SSH and admin rights. The steps assume a MacRent Mac Mini, but any Mac works.
• A model provider: a Nous Portal account, an OpenRouter key, or a local model through Ollama.
• A Telegram account to create the bot with, or whichever messaging app you prefer.

Step 1: Install Hermes over SSH

Connect with ssh username@your-mac-ip from the welcome email, then run the install script from the Hermes documentation: curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash. It installs uv and Python 3.11, clones the repository and needs no sudo. Open a new shell and run hermes setup for the interactive wizard, then hermes model to choose between Nous Portal (OAuth), OpenRouter (API key) or a custom endpoint.

Step 2: Try it in the terminal

Run hermes. That is the full interactive CLI with tools, memory and skills. Give it a small real task so you see how it asks for permissions and where it writes its files. Everything it remembers goes under ~/.hermes/.

Step 3: Connect Telegram and run the gateway

Create a bot with BotFather in Telegram and copy its token. On the Mac, run hermes gateway setup; the wizard walks through Telegram, Discord, Slack and WhatsApp. Then start the gateway with hermes gateway and message your bot from your phone.

The documented hermes gateway install registers a systemd service, which macOS does not have. On a Mac, either keep the gateway in a tmux session (tmux new -s hermes, run hermes gateway, detach with Ctrl-b d), or write a small LaunchAgent plist in ~/Library/LaunchAgents that runs the same command with KeepAlive set, and load it with launchctl. A MacRent Mac never sleeps, so either survives you closing your laptop.

Step 4: Optional local model

brew install ollama, start it, and ollama pull a 7B to 9B model. Run hermes model again and point it at the custom endpoint on localhost. A 16GB Mac Mini runs that size at conversational speed with Metal; keep the cloud model configured for the hard tasks.

Keeping it healthy

• Updates: hermes update pulls the latest code and reinstalls dependencies.
• Secrets: keep only the tokens the agent needs on the box, scoped as narrowly as the provider allows.
• Backups: copy ~/.hermes/ somewhere private on a schedule. The skills it writes for itself are the part you will miss.
• Reset: if the agent breaks the Mac, cancel and rent a fresh one, or email us for a reset, then restore the directory.

What it costs

A dedicated Mac Mini on MacRent is $15 for a day, $89 for a week or $159 for a month, on around the clock with a static IP. Your Nous Portal or OpenRouter usage is separate. If you are still choosing between the two agents, the OpenClaw on a cloud Mac page covers the same setup on the other side.