Video: "Agent OS: Local Setups, VPS & Higgsfield!" by Julian Goldie on YouTube.
What this video covers and why the question matters
Agent OS is Julian Goldie's term for the broader stack around Hermes Agent — the dashboard, memory system, voice controls, and tool integrations that turn a single AI agent into something closer to an operating system for your work. Once you have that stack built, the obvious next question is how to keep it running when you are not at your desk.
This video tackles that directly. It is not a beginner introduction to Hermes — it assumes you already have an Agent OS running and are asking what comes next. The three answers Julian works through are: run it locally and use Tailscale to access it from anywhere, host it on a VPS so it runs even when your laptop is off, and add the Higgsfield MCP so the agent can generate visual content without you leaving the workflow.
Local setup: Tailscale puts your Agent OS in your pocket
The common assumption about local AI is that you are tied to your desk. Tailscale breaks that. It is a free private networking tool that creates an encrypted tunnel between your devices — install it on your computer and your phone, and they share a private HTTPS address that nothing external can reach. No open ports, no router changes, no public URL.
The practical result is that you can open your phone, toggle Tailscale on, type the private address, and get a full view of your Agent OS — Kanban board, agent chat, memory, active jobs — from anywhere. Julian says setup takes around ten minutes and Claude can handle most of it from a single prompt. The main limitation is what it says on the tin: your local machine has to be on and connected for this to work. If your laptop goes to sleep, your agents stop.
Worth knowing: the Tailscale free tier is enough for personal and small business use. You are not paying for this.
VPS hosting: when always-on is worth the monthly cost
If your agents are doing scheduled work — monitoring keyword feeds, sending reports on a timer, processing incoming emails — they cannot afford to stop when you close the laptop. That is the case for a VPS. Julian's example uses Hostinger, where a suitable VPS sits at a few pounds per month. You install the same Agent OS stack you run locally, and it runs continuously with no dependence on your home machine.
The tradeoff is straightforward: you are paying a fixed monthly cost in exchange for reliability and always-on access. For agents that do meaningful background work — the kind that earns back the cost in time saved — a VPS is the sensible choice. For an agent you mostly interact with directly, local plus Tailscale is cheaper and simpler. Julian does not argue for one over the other; he presents both and lets the use case decide.
Higgsfield MCP: adding a film studio to your agent
The third section of the video is the most visually striking. Higgsfield is an AI image and video platform with over 30 models — including Kling, Seedream, Veo, Soul, and Cinema Studio — all accessible through a single interface. The Higgsfield MCP is a connector that plugs all of that into any MCP-compatible agent, including Hermes, Claude, and OpenClaw.
Once it is wired in, you describe what you want in plain English and the agent picks the right model, writes the prompt, submits it, and files the output in your workspace. Generation takes roughly 30 to 60 seconds. The output is 4K images or cinematic video depending on what you asked for. The agent handles the whole sequence — you are not switching tabs or logging into Higgsfield separately.
In practice this means a content workflow that previously required you to open Higgsfield, write a prompt, download the output, and bring it back into whatever you were doing becomes a single instruction to the agent. Useful for regular visual content — product imagery, short clips for social, ad creative — where the manual process adds up. The quality ceiling on Higgsfield's models is genuinely high; the MCP does not reduce it.
What is actually new here
None of these three tools — Tailscale, VPS hosting, Higgsfield — are brand new. What is new is the framing: Julian is treating Agent OS as infrastructure that needs proper deployment thinking, not just a desktop tool you switch on when you need it. That is a meaningful shift. A year ago the conversation about Hermes was mostly "what can it do?" Now it is "how do you run it properly so it actually does those things reliably?"
That is a sign of maturity in the tooling, and it is worth paying attention to if you are building anything serious on top of these agents. The Tailscale approach in particular is underused — most people either give up on local and go full VPS, or they run local and lose access the moment they step out. The middle path is cheap and practical.
Where this connects to NordSys
Helping businesses deploy AI agents properly is exactly what our AI Agents service covers — whether that is configuring a local Agent OS with mobile access, moving an existing setup to a VPS, or integrating creative tools like Higgsfield into an automated content workflow. If you have a Hermes Agent or Claude Code setup that is not running reliably, or you want to add visual content generation to an existing agent workflow, our AI Agents service is the right starting point.
See our AI Agents →