Agent OS is the layer I get asked about more than anything else in my community, and most people want to see the inside of mine before they build their own. So here's the full walkthrough. Every layer, every panel, every reason I built the stack the way I did.
This is the inside-the-stack post. I'll walk you through mission control panel by panel, the Goldie Mission Stack layer by layer, and the specific design decisions that made the stack worth building. Treat this as the deepest read I've published on the agent os so far.
Want the same Agent OS on your machine? The full zip lives inside AI Profit Boardroom with 100+ prompts and the 30-day roadmap. $59/mo locked forever, twin guarantee, 5 weekly coaching calls. Get the stack
Why I Call It An Agent OS Rather Than An AI Workflow
Before the walkthrough, the naming matters. I deliberately call this an agent os rather than an AI workflow or an AI stack, because the difference is the whole point.
A workflow is a sequence of prompts and tool calls for one task.
A stack is the set of tools you happen to use.
An operating system is the thing that coordinates everything else.
iOS doesn't make calls. iOS doesn't write messages. iOS doesn't take photos.
What iOS does is coordinate every app on the phone into one coherent system with shared contacts, shared notifications, shared file system and shared identity.
That's the job of an agent os. It coordinates Claude, Hermes, OpenClaw and the Self layer into one personal operating system that runs my business every day.
I've explained the same naming choice in hermes agent os and what is agent os for anyone who wants the wider framing.
The Hammer Vs Construction Company Frame
The other thing I want to lock in before the walkthrough is the leverage frame.
Using Claude on its own is owning a hammer.
Running an agent os is running a construction company.
Both build things. But the hammer needs you to swing it every single time, and the construction company keeps building when you step away.
Day one of using a hammer looks similar to day one of running a construction company.
Day thirty looks completely different.
Keep that frame active while you read the walkthrough. Every panel I describe is either building the construction company or it's a hammer dressed up in Tailwind.
The Mission Control Front Door
Mission control is the first screen I see in the morning. The layout has stayed almost identical since I shipped the first version because the shape just works.
Left rail: live status indicators for every agent in my stack. Claude. Hermes. OpenClaw. Plus any custom agent I've wired in.
Centre column: the active chat with whichever agent I'm driving right now. Full chat history scrolls up.
Right rail: goals tracker with progress bars across my key projects.
Below the chat: daily journal section that captures what got built and what got blocked.
Top right: quick access to per-agent control rooms.
The whole front door is one screen. No tabs. No app switching. One operating system surface, the way iOS gives you one home screen.
I covered the dashboard surface separately in hermes agent mission control if you want the visual breakdown.
Layer 1 — Intelligence (Claude / Claude Code)
The Intelligence layer is where every conversation starts. Claude and Claude Code sit at the top of the stack as the CEO.
Claude plans the work.
Claude decides what gets prioritised.
Claude Code executes the actual builds when something needs to ship.
Inside mission control, Claude has the primary chat surface. Every other agent runs in support of what Claude is planning.
This is the agent I spend most of my driving time on, and it's the one most people think IS their AI stack. The whole point of the agent os is that Intelligence is only one of four layers.
The deeper Claude integration is documented in claude hermes agent.
Layer 2 — Execution (OpenClaw)
The Execution layer is OpenClaw. It's the local gateway that turns a single-agent setup into a multi-agent team.
OpenClaw routes tasks between agents.
OpenClaw manages sessions across the stack.
OpenClaw coordinates the actual work between Claude and Hermes.
Inside mission control, OpenClaw gets its own control room with session history, routing config and the API keys for the agents it manages.
If Claude is the CEO, OpenClaw is the COO. It takes what Claude decides and parcels execution out to the right specialist.
The deeper OpenClaw breakdown is in openclaw computer use for anyone who wants the full picture.
Layer 3 — Research (Hermes)
The Research layer is Hermes. This is the workhorse that runs the long-running multi-step jobs.
Hermes runs the tool calls.
Hermes runs the Kanban-driven workflows.
Hermes manages skills and plugins.
Hermes handles browser automations and any operational job that would melt a single Claude session.
Inside mission control, Hermes has its own control room with the Kanban board, the skills and plugins list, the tool-call log and the session history.
This is the layer where the heavy lifting happens once Claude has decided what needs doing. The install walkthrough is at hermes agent installation guide 2026 and the framework view is in hermes ai agent framework 2026.
Layer 4 — Self (Obsidian Vault + OMI)
The Self layer is the layer almost nobody talks about and the layer that makes the whole agent os actually feel personal.
OMI records what's happening on my screen and through my mic during the working day.
OMI exports the transcripts to my Obsidian vault overnight.
Every agent in the stack pulls personal context from that vault on every prompt.
Inside mission control, the Self layer surfaces as a "context" panel that shows which vault notes were pulled for the current conversation.
This is the unlock. Context is the single biggest driver of AI performance. Without the Self layer, your agents produce generic output. With it, the output is specific to your business, your customers, your projects and your voice.
I've written about the Obsidian side in claude obsidian setup for anyone setting up the vault path.
The Per-Agent Control Room
The control room is where you can really tell this is an operating system rather than a chatbot wrapper. Every agent in the stack has its own dedicated panel with the same set of tabs.
API keys and provider config so you can swap models without breaking anything else.
Session history with full chat logs.
Skills and plugins specific to that agent.
A Kanban board for the work that agent is queued up to do.
Insights and analytics scoped to that agent — sessions, tool calls, tokens, peak hours.
Same shape across Claude, Hermes and OpenClaw. Consistent UI, consistent shortcuts, consistent expectations.
That consistency is the difference between an operating system and a folder of scripts. iOS apps all have the same gestures and the same shape. The control rooms in mission control follow the same principle.
The Analytics View
The analytics view is where I check whether my stack is actually paying for itself. Without analytics, you can't optimise. With analytics, you can see exactly where the time and tokens are going.
Sessions per agent per day.
Tool calls per agent per day.
Token consumption per agent per model.
Peak hours chart so I can see when I'm actually using the system.
Daily journal entries that capture what got built and what got blocked.
This isn't vanity dashboards. The analytics let me move spend toward the agents that are doing the most useful work, and pull spend off the agents that are sitting idle.
It also gives me an honest signal of when I'm actually using AI to ship rather than to fiddle.
Why I Built It Locally
The whole agent os runs locally on my Mac. Not on a server, not in the cloud beyond the model APIs themselves. Three reasons.
The Self layer holds personal data. That can't live on someone else's server.
Latency drops to essentially zero on local routing between agents.
The stack keeps working if my wifi drops or a provider has a bad day.
I wouldn't run an agent os any other way. The cloud-hosted equivalents are fine as products, but they lose the personal context piece that makes the stack actually feel like mine.
The broader local-first case is in claude code local for anyone weighing it.
How I Built The First Version In One Hour
The build story is worth telling because it's faster than people expect. The first working version of my agent os shipped in roughly one hour using Claude Desktop.
I opened Claude Desktop and described the four-layer Goldie Mission Stack in detail.
I pasted in the Hermes and OpenClaw documentation from GitHub.
I asked Claude to scaffold the dashboard in Next.js and Tailwind.
I ran the result locally, captured the first errors, pasted them back to Claude, and asked for fixes.
By the end of the hour, mission control was live on localhost.
The full build walkthrough is over in the companion post on agent os claude for anyone who wants to follow the same path. The Claude Desktop side of the wiring is also in claude hermes agent.
What A Day Inside The Agent OS Actually Looks Like
For the people who haven't seen this run yet, here's what an actual working day inside my agent os looks like.
I open mission control as the first app of the day.
The morning panel shows yesterday's journal entry and the queued goals.
I drive Claude in the centre column to plan the day's priorities.
Claude routes the execution work through OpenClaw.
OpenClaw kicks off two or three Hermes workflows in the background — research jobs, draft pulls, Kanban movements.
I work on something else while Hermes runs.
Mid-morning, the Hermes jobs land back in mission control and I review the output.
Afternoon, I drive Claude Code to ship one or two builds that came out of the morning's planning.
End of day, the agent os auto-writes the journal entry to Obsidian.
Overnight, OMI export feeds the day's transcripts back into the vault for tomorrow's context.
The whole thing runs as one continuous business operating system rather than four disconnected apps. That's the leverage. That's the agent os.
What This Stack Replaced In My Workflow
The honest version of what changed when I started running this stack.
Five disconnected tabs collapsed into one mission control.
Daily re-pasting of context collapsed into automatic context pull from the vault.
Sequential single-agent chats collapsed into parallel multi-agent workflows.
Manual copy-paste between Claude and Hermes collapsed into automatic OpenClaw routing.
No memory of yesterday collapsed into full continuity carried overnight.
Generic output collapsed into output that sounds like me, references my customers, and slots into my existing content strategy.
That's not a marginal upgrade over disconnected AI tabs. That's a different category of tool entirely.
Inside AIPB — The Full Agent OS Bonus Stack
If you want this same stack running on your machine, the whole thing is bonused inside AI Profit Boardroom at $59/mo locked forever.
What's in the Agent OS bonus pack:
The full Agent OS zip ready to install on your machine.
100+ Agent OS prompts I use across Claude, Hermes and OpenClaw inside mission control.
The 30-day roadmap from install to fully operational mission control.
The broader Boardroom — 5 weekly coaching calls, 3,000+ members building these systems, daily Q&A with me, 1,000+ done-for-you AI workflows, and a 7-day refund plus 30-day ROI twin guarantee.
This is the exact stack I run my Goldie Agency on. Not a theory build. The actual system.
Get my full Agent OS on your machine Join the AI Profit Boardroom at $59/mo locked forever and grab the Agent OS zip, 100 prompts, 30-day roadmap, plus weekly coaching with me. Get inside
The AIPB Walkthrough — See What's Inside
If you want to see the Boardroom before you join, here's the walkthrough. It covers the calendar of weekly calls, the bonus stack including the Agent OS pack, and the community where members ship these builds together.
The walkthrough will show you why I'd recommend this as the on-ramp to a real agent os.
Free AI Money Lab — Start Without Paying
If $59/mo isn't where you're at yet, the AI Money Lab is my completely free community. Public training, a slice of the prompt library, and a slower walk through the Goldie Mission Stack.
The right on-ramp if you want to build your own agent os from scratch without the bonus pack.
Strategy Session — Custom Builds From Goldie Agency
For business owners who want my team to build a custom agent os around their company, I take a small number of strategy calls through the Goldie Agency. Book free at go.juliangoldie.com/strategy-session.
The path for agencies, SaaS founders and operators who want it custom-fitted rather than templated.
FAQ — Inside Julian's Agent OS
What does your agent os actually run on, hardware-wise?
It runs on my Mac. Any modern Mac handles the dashboard, Hermes and OpenClaw fine. The heavy compute happens on the model API side rather than locally.
How big is the Agent OS zip?
The install is lightweight because it's a Next.js scaffold plus the integration layer. The vault and OMI exports are the parts that grow over time — those live in your Obsidian folder separately.
Why is the Self layer so central to your version?
Because context is the single biggest driver of AI output quality. Without the Self layer pulling from my Obsidian vault and OMI transcripts, my agents produce output that any other user could produce. With it, the output is specifically mine.
Do you really use this every day?
Yes. Mission control is the first app I open every morning and one of the last things I check at night. It's the operating system for my Goldie Agency work, my Boardroom content, and my YouTube production stack.
What's the hardest part of replicating this stack?
The Self layer. Wiring the Obsidian vault and OMI export paths takes the most thought because it's the most personal piece of the build. The bonus pack inside AI Profit Boardroom ships templates that get you 80 percent of the way there.
Can I extend the stack with my own agents?
Yes. The dashboard is built to add new agents via the per-agent control room pattern. Any tool with an API surface can be wired in alongside Claude, Hermes and OpenClaw.
About Julian
I'm Julian Goldie — AI entrepreneur, SEO expert and founder of the AI Profit Boardroom with 3,000+ members. I run Goldie Agency, a 7-figure SEO and AI agency, and I've published "SEO Link Building Mastery" and "Agency Marketing Mastery" on Amazon.
I help business owners scale with AI agents, automation, and the agent os stack I run on my own machine every day.
Get my best AI training inside the AI Profit Boardroom
Also On Our Network
- Read on bestaiagentcommunity.com
- Read on aiprofitboardroom.com
- Read on juliangoldieaiautomation.com
- Read on aisuccesslabjuliangoldie.com
Related reading
- Hermes Agent OS for founders — the founder-focused view.
- Agentic AI OS — the broader architecture piece.
- Hermes Agent Mission Control — the dashboard up close.
- Claude Obsidian Setup — the Self layer in detail.
That's the inside view of my agent os, and it's the stack I'd build first if I were starting over today.