Agentic OS Claude is the project I built from one prompt in roughly an hour, and this post is the exact build guide for replicating it on your own machine.
The crazy part is that the brain of the OS built the OS itself.
I gave Claude Desktop one paragraph and watched it scaffold the whole dashboard with no hand-holding.
This is the exact step-by-step I used to ship my own Agentic OS Claude setup in one sitting, including the prompt, the wiring, and the panels you'll want.
Want the full Agentic OS Claude build files? Inside the AI Profit Boardroom, I share the prompt templates, the layout JSON, and the CLI bridge code. Five weekly calls, 3,000+ members, $69/mo locked with the twin guarantee. Get access here
Before You Start — What You'll Need
The Agentic OS Claude build needs three things on your machine before you begin.
You need Claude Desktop installed with at least filesystem and terminal MCPs connected.
You need Node.js installed because Claude is going to spin up a Next.js project for the dashboard.
You need an Obsidian vault folder ready because that's where the OS will write its memory log.
That's the entire prerequisite list.
If you have those three, you can finish the build in about an hour.
For the higher-level context on why we're building this, see my agentic OS post.
Step One — Open Claude Desktop And Confirm MCP Access
The first move is to make sure Claude Desktop can actually do things on your machine.
Open the MCP settings panel and confirm filesystem read/write is enabled.
Confirm at least one terminal MCP is wired so Claude can run shell commands.
If you have a browser MCP installed, enable that too — it makes the dashboard build smoother.
Without those MCPs, Claude can talk about building an OS but can't actually build one.
This is the cheapest five-minute setup you'll do all day.
Step Two — Drop The One-Prompt Build Brief
This is the prompt I used to scaffold the whole Agentic OS Claude project.
I want you to use it verbatim because the phrasing is half the trick.
Create a beautiful operating system hosted locally for managing
Claude for a website connected to Claude. Should be like a beautiful
mission control dashboard. Then allow me to control my OpenClaw, my
Hermes, and any other agents in separate systems inside the dashboard.
Paste that into a fresh Claude Desktop chat and hit send.
Claude will come back with a few clarifying questions about stack, panels, and styling.
Tell it Next.js plus Tailwind CSS for the stack, four panels for the layout, and clean minimal styling.
Tell it to host locally on a free port and stream agent logs into the dashboard.
That's enough information for Claude to take over.
Step Three — Watch Claude Scaffold The Project
This step is mostly you sitting back while Claude does the work.
Claude will create the project folder, run npm init, install Next.js, set up Tailwind, and write the initial component tree.
It will create the four panel layout with placeholders for Intelligence, Execution, Research, and Self.
It will spin up a small backend API folder for the bridges to downstream agents.
It will run npm run dev and tell you which localhost port the dashboard is now serving on.
If anything errors, paste the error back into the same chat and let Claude fix it.
That's the magic — Claude self-heals its own build because it's still in the same conversation.
You can read my agentic OS Claude Code post for the longer Claude Code variant of this build.
Step Four — Wire In The Claude CLI Bridge
This is the step that separates a real Agentic OS Claude from a glorified chat tab.
The Claude CLI bridge lets the Intelligence panel actually call Claude with full tool access on your machine.
Without the bridge, the Intelligence panel is a fancy chat window.
With the bridge, the Intelligence panel becomes a real command surface for Claude with filesystem, terminal, and MCP access.
Tell Claude in the same chat: "Add a Claude CLI bridge so the Intelligence panel spawns a Claude subprocess with full tool access on every prompt".
Claude will write the bridge code, plumb it through the API folder, and update the Intelligence panel to call it.
Restart the dev server and confirm the Intelligence panel now responds with real Claude output, not mock data.
That's the bridge done.
Step Five — Add The OpenClaw Execution Panel
The Execution panel is for browser tasks and visual work.
OpenClaw is the agent I run there because it's reliable and fits cleanly into the OS pattern.
Tell Claude in the same chat: "Wire the Execution panel to dispatch tasks to my OpenClaw endpoint at localhost:port".
Claude will update the Execution panel UI, wire the API call, and add a small queue display so you can see tasks being processed.
Test it by typing "open google.com and screenshot the homepage" into the panel and watching OpenClaw do the work.
That's the Execution layer alive.
For a deeper look at how OpenClaw fits the stack, my Claude + Hermes agent post shows the full pairing in action.
Step Six — Add The Hermes Research Panel
The Research panel is for long lookups and multi-step tool chains.
Hermes is the agent that lives here because it handles deep investigations without polluting the main Claude thread.
Tell Claude: "Wire the Research panel to dispatch research tasks to my Hermes endpoint with a streaming progress display".
Claude updates the Research panel UI, wires the API call, and adds a streaming feed so you can watch Hermes work.
Test it with "research the top three AI agent frameworks in 2026 and summarise" and watch the panel populate.
That's the Research layer alive.
Step Seven — Wire The Self Panel To Obsidian + OMI
The Self panel is the memory layer, and it's the bit that turns this from a single-session toy into a real OS.
Tell Claude: "Wire the Self panel to my Obsidian vault at /path/to/vault and display the most recent ten notes with a search box".
Claude will read your vault path, wire a file watcher, and update the panel to show recent notes.
If you have OMI captures piping into the vault, add: "Display OMI-tagged notes separately with a timestamp".
Test by writing a note in Obsidian and watching it appear in the Self panel within seconds.
That's the memory layer alive.
I cover the daily routine with the Self panel in my agentic OS command center post.
Step Eight — Add Auto-Logging Of Every Claude Chat
This is the optional step that makes the OS feel alive.
Tell Claude: "Auto-log every Intelligence panel chat into a daily note in the Obsidian vault with timestamps".
Claude updates the bridge to append to a daily markdown file in the vault every time you send a prompt.
Now every Claude session, every OpenClaw task, every Hermes report ends up in the vault automatically.
You never have to copy-paste a chat log again.
Tomorrow's first task is to ask Claude "what did we do yesterday?" and watch it read the vault and give you the brief.
That's the loop that makes the OS compound over time.
The Goldie Mission Stack Inside Your Build
Once you've finished the seven steps above, you have what I call the Goldie Mission Stack running on your machine.
The stack has four layers and Claude sits at the centre of all four.
Layer 1 — Intelligence (Claude + Claude Code).
The brain — plans and dispatches.
Layer 2 — Execution (OpenClaw).
The hands — browser tasks.
Layer 3 — Research (Hermes).
The legwork — long lookups.
Layer 4 — Self (Obsidian + OMI).
The memory — vault logs and voice notes.
That's the same stack I use across every project, every day, in my agency and my own work.
It's deliberately simple because complex stacks fail under real-world pressure.
For the original naming convention of the same stack, see my agent OS Claude post.
Where Most People Get Stuck
I want to flag the common stuck points so you can skip past them.
The most common one is forgetting to enable MCP access in Claude Desktop before starting the build.
Without MCP access, Claude can talk about building an OS but can't actually run any commands.
The second one is skipping the Claude CLI bridge.
Without the bridge, the dashboard is decorative.
With it, the dashboard is operational.
The third one is not pointing the Self panel at a real Obsidian vault.
Without the vault, the OS has no memory and forgets everything between sessions.
With the vault, the OS compounds intelligence every single day.
Fix those three and you're already ahead of 95% of people trying to build their own Agentic OS Claude setup.
What You Can Run Through The OS Right Away
Once the build is finished, here are real tasks I run through mine on day one.
I ask Claude to plan my content week from the vault notes that landed overnight.
I dispatch Hermes to research three competitor pricing pages and summarise the findings.
I send OpenClaw to log into my CMS, post a draft, and screenshot the result.
I ask Claude Code to refactor an Eleventy template and ship the update.
I log everything to the vault so tomorrow's session has the full context.
Five jobs that used to need five tools now happen in one dashboard with Claude routing them.
The agentic OS download post has the build pack contents if you want to extend the OS further.
How Long Each Step Actually Takes
For honesty's sake, here's a realistic time breakdown if you're following along.
| Step | What you do | Realistic time |
|---|---|---|
| 1 | Confirm Claude Desktop MCPs | 5 minutes |
| 2 | Drop the build prompt | 2 minutes |
| 3 | Watch Claude scaffold the project | 15-20 minutes |
| 4 | Add the Claude CLI bridge | 10 minutes |
| 5 | Add the OpenClaw Execution panel | 10 minutes |
| 6 | Add the Hermes Research panel | 10 minutes |
| 7 | Wire the Self panel to Obsidian | 5 minutes |
| 8 | Add auto-logging to the vault | 5 minutes |
That's about 60-65 minutes total if you don't get stuck on anything.
If you do get stuck, paste the error into the same Claude Desktop chat and let Claude debug itself.
That's another underrated advantage of having the brain build the OS — it remembers everything it just shipped and can fix it in context.
What I Learned From Building It
I learned three things from this build that I want to pass on.
The first is that one good prompt beats ten mediocre ones.
The shorter and clearer the brief, the more sensibly Claude builds.
The second is that the brain has to be Claude.
I tried other models in the centre seat and the build kept stalling on multi-step planning.
The third is that the vault is the unlock.
Without persistent memory, every session starts from zero and the OS feels like a toy.
With the vault, every session compounds on the last and the OS becomes a real operating system.
That's it — three lessons that took me a few weekends of trial and error to figure out.
For the wider meaning and category context, see my agentic OS meaning post.
Watch The Full Build Walkthrough
Here's the community space where I share the full build walkthrough with screen recordings.
Five weekly calls, 3,000+ members, $69/mo locked with the twin guarantee.
That's the room where I do live builds of the Agentic OS Claude setup with members.
FAQ — Agentic OS Claude Build
Can I really build an Agentic OS Claude in one hour?
Yes, the eight-step build I described above takes roughly 60-65 minutes if your prerequisites are in place and you don't hit any blocking errors.
What if Claude's scaffolded build has bugs?
Paste the error back into the same Claude Desktop chat and Claude will self-heal the build because it still has the full context of what it just shipped.
Do I need to know Next.js or React to follow this build?
No, Claude writes all the code itself — you just need to be able to read terminal output and tell Claude where to point the bridges.
Can I swap OpenClaw or Hermes for other agents?
Yes, the Execution and Research panels are agent-agnostic — you can wire any downstream tool that exposes a sane API endpoint.
What's the smallest version of Agentic OS Claude I can ship?
The smallest viable build is just steps one through four — Claude Desktop, the prompt, the scaffold, and the CLI bridge. That alone is more useful than a plain chat window.
Where do I get the full Agentic OS Claude build pack?
The full build pack with the prompt variations, the bridge code, and the panel templates lives inside the AI Profit Boardroom at $69/mo locked.
About Julian
I'm Julian Goldie — AI entrepreneur, SEO expert, and founder of the AI Profit Boardroom (3,000+ members at $69/mo locked with twin guarantee).
I run Goldie Agency, host five weekly coaching calls inside the Boardroom, and have authored multiple books on SEO and AI automation.
- 7-figure AI agency (Goldie Agency)
- Five weekly coaching calls inside the Boardroom
- Author of "SEO Link Building Mastery" and "Agency Marketing Mastery"
- 50,000+ Udemy students across long-running courses
→ Get my best AI training inside the AI Profit Boardroom
Further Reading On Agent OS Guide
For deeper walkthroughs on the topics in this article, the Agent OS Guide library has these worth bookmarking.
- Agent OS Guide — the full library if you want every Claude-driven agentic os angle in one place.
- How Claude Mythos Changed AI — context on why Claude sits at the centre of so many agentic os builds.
- Free Claude Code Setup Guide — a clean walkthrough for wiring Claude Code into the agentic os.
- Hermes Agent OS Q&A — answers the questions that come up when Hermes joins a Claude-first stack.
Also On Our Network
- Read on bestaiagentcommunity.com
- Read on aiprofitboardroom.com
- Read on aisuccesslabjuliangoldie.com
- Read on aimoneylabjuliangoldie.com
Related Reading
- Agentic OS — the bigger picture explained
- Agentic OS meaning — what the term really means
- Agentic OS Claude Code build guide
- Agentic OS command center daily routine
- Agentic OS download — the build pack
- Agent OS Claude original walkthrough
- Claude + Hermes agent pairing
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
That's the one-hour build path for your own Agentic OS Claude setup in 2026.
