Hermes on my desktop: local models, frontier APIs, dashboard on my phone

I run Hermes Agent on the PC in my office.

Hermes is an open-source AI agent from Nous Research. Tools, memory, skills, a terminal, messaging hooks, and a real web dashboard. Not just another chat tab.

Chatbots talk. Agents act.

A chatbot mostly returns text. You type, it replies. Useful, but limited.

An agent can take steps. It plans, calls tools (files, terminal, browser, APIs), checks what happened, and keeps going until the job is done or it needs you. Chat is the interface. Tools and the loop are the difference.

Running Hermes locally means those hands sit on my machine — my files, my shell, my network — not a sealed web chat that only spits out paragraphs.

Two ways to feed it brains

I wired it up two ways:

• Ollama for local models on this machine
• API keys for frontier models (Anthropic, xAI, and the rest) when I want more power

Same agent either way. Local when I want work on-box and cheaper tokens. API when the job needs a stronger model and I’m fine paying for it.

Local is slower if you’re used to frontier speed. That’s real. You wait more. For a lot of everyday stuff it’s still useful, and it’s a fun project to run the stack yourself instead of only renting someone else’s chat box.

The part that made it click: Tailscale + the dashboard

Chat through iMessage or Telegram is fine for short asks. It’s not the full product.

Hermes has a real dashboard — sessions, settings, tools, the whole suite. I wanted that on my phone, not only a text thread.

So I installed Tailscale on the office desktop and on my iPhone. Both join a private mesh network. No opening ports on the router. No hanging the dashboard on the public internet.

From my phone’s browser I hit the machine running Hermes and get the full dashboard. Same controls I’d use at the desk, in my pocket.

That matters more than another messaging integration. Full access, on hardware I own.

What I have now

• Hermes on my desktop
• Ollama when local is enough
• Frontier APIs when it isn’t
• Tailscale so the phone can reach the box directly
• The real dashboard, not just chat

It all lives on my machine. I’m not dumping everything into a random web app I don’t run.

Where it could go

If local gets too slow, I can upgrade the GPU in that box. If I want it always on without the desk PC, I can deploy Hermes in the cloud. Either way, Tailscale still lets me connect straight from my phone the same way.

Presence and horsepower can move. The access pattern stays simple: my devices, private network, full UI.

Why bother

I run a one-person tech shop in northeast Nebraska. I like tools I control. This setup isn’t magic, and local models will test your patience if you live on frontier speed all day. It’s still a solid way to run a real agent at home, save tokens when you can, and keep the good interface with you when you leave the office.

Hermes: hermes-agent.nousresearch.com