Video: "Agent OS: Claude + Hermes AI = Superpowers!" by Julian Goldie on YouTube.
What the Agent OS actually is
For the last few months, people running Claude, Hermes and OpenClaw have been switching between three separate interfaces and manually copying context between them. That friction is the problem the Agent OS addresses.
Julian's build is a locally hosted dashboard — built in Next.js and Tailwind — that puts chat panels, a goals tracker, a journal, a memory layer and per-agent control rooms onto one screen. It runs on your Mac, handles all three tools through a shared interface, and keeps them pointed at the same goals using the same memory. Nothing is sent to a third-party cloud. The whole build took him about an hour.
The four-layer stack
The system has a clear division of labour. Claude handles the thinking and writing — it's the intelligence layer, the one you prompt with a goal and expect a reasoned output from. OpenClaw handles execution: it can click, type, scroll and navigate your desktop apps on your behalf. Hermes handles research and multi-modal production — images, video, voice — with Grok plugged in for live data.
The fourth layer is memory: Obsidian stores a knowledge base of your goals, journal entries and conversation history, while the OMI wearable captures voice signals during the day. Hermes reads that vault before each task so it starts with context rather than from scratch.
In practice this means you set a goal once — say, a week's content campaign — and each agent draws from the same brief, tracks progress on the same board, and builds on each other's outputs rather than duplicating them.
What it changes for a business owner
The obvious win is that you stop re-explaining context. Every tool in the stack knows what the others have done and what the goal still is. That sounds like a small convenience but it compounds quickly across a week of work.
Worth knowing: setting this up is not plug-and-play. You still need to install and configure Hermes and OpenClaw, connect the Obsidian memory vault, and build or adapt the dashboard. Julian built his in an hour, but he's been working with these tools daily for months. A business owner starting from scratch should expect closer to a day — or hire someone who already knows the stack.
The more interesting implication is what this pattern represents. Agent OS isn't a product you buy — it's an architecture. The idea that your AI tools should share memory and goals rather than running as isolated chatbots is now practically achievable with free, open-source components. That wasn't true twelve months ago.
What's still rough
OpenClaw's computer use feature is powerful but not always reliable on complex desktop tasks — it can misfire on windows it hasn't seen before. The memory layer needs regular pruning or it accumulates noise. And the dashboard Julian built is a prototype, not a polished product: it works but you'll want to adapt it.
That said, the combination of Claude's reasoning, Hermes's persistent memory, and OpenClaw's ability to operate your machine is genuinely a step change from running any one of them alone.
Where this connects to NordSys
Setting up Hermes Agent, Claude Code and OpenClaw as a coordinated stack — configured properly for your specific work — is exactly what we do for clients. If you want the Agent OS architecture without spending a week figuring out the plumbing, that's a practical conversation to have.
See our AI Agents service →