AI Operating System — your team's single pane of glass, built with Claude Code

Download this repo, hand it to Claude Code, and tell it what to change. The layout, the tabs, the integrations, the skills, the wiki schema — all of it is yours to reshape. AI OS is the working skeleton so you don't start from a blank file.

Out of the box you get a real, deployable Next.js dashboard with Claude tool-use against Stripe, HubSpot, and Google Search Console. The point isn't that it's done — the point is that it's real enough for Claude Code to mold it into the single pane of glass your team actually needs.

Not a developer? Good. The workflow is: clone, open in Claude Code, describe what you want changed in plain English. Claude Code edits the files for you. You don't need to learn React.

Built by Mark Fershteyn — founder of MasteringAI, where executives and operators learn to build with AI instead of just using it. Don't want to self-host? Zealos is the premium "built for you" version.

---

How people actually use this

Three workflows, all starting with "give it to Claude Code":

1. Reshape the layout. "Claude, replace the Skills tab with a Pricing Experiments tab that pulls from Stripe and shows ARPU by cohort." → Claude edits the nav, builds the page, wires the data.

2. Swap integrations. "Claude, remove HubSpot and add Salesforce instead, mirroring the same patterns." → Claude follows the existing integration scaffold and ports it.

3. Rewrite for your role. "Claude, this is for a Head of Product, not a CRO. Rewrite the skills catalog and wiki schema for product strategy." → Claude updates data/skills/ and data/wiki-schemas/.

You're not learning a framework. You're using English to direct an AI agent against a working codebase.

Who this is for

• Executives & operators who want their own internal dashboard without hiring an engineer.
• Non-technical builders using Claude Code who need a real, production-shaped repo to direct — not a tutorial toy.
• Teams building internal tools who'd rather start from a working OS than a blank Next.js page.

What's already wired up

So Claude Code has good bones to remix:

• Anthropic tool-use against real APIs — Stripe, HubSpot, Google Search Console. Copy the pattern for any other API.
• An AI-maintained wiki so chat doesn't re-fetch raw data every turn. (Inspired by Karpathy's LLM Wiki pattern.)
• BYOK auth — per-user API keys, encrypted at rest. Works for solo or team deploys.
• Skills catalog — 30 starter "executive plays" (board memo, pipeline briefing, competitor teardown, 1:1 prep, inbox triage…). Each is a markdown file Claude Code can rewrite for your role in one prompt.
• Two run modes — personal (single-user, local JSON store) or multi-user (Google OAuth + Supabase).
• Next.js 16 where every page reads real data, no mock fixtures — so Claude Code edits flow straight to production.

Quick Start

Run these four commands in your Terminal app. If you've never used Terminal, search "how to open terminal on Mac/Windows" first — takes 30 seconds.

git clone https://github.com/Fersh/ai-operating-system.git
cd ai-operating-system
cp .env.example .env.local
echo "AI_OS_ENCRYPTION_KEY=$(openssl rand -base64 32)" >> .env.local
npm install
npm run dev

Open http://localhost:3000, click Settings, paste your Anthropic API key (grab one at console.anthropic.com/settings/keys), then connect Stripe, HubSpot, or Google Search Console on the Integrations page.

Now make it yours

Open the project folder in Claude Code and start prompting. Openers that work:

• "Walk me through the file structure so I know what's where."
• "Replace the Skills page with a Pricing page that uses Stripe to chart MRR by plan."
• "Add a dark mode toggle in the header."
• "Add an integration card for Linear — I'll paste my API key once it's wired."
• "Change the brand color from charcoal to deep green throughout the app."

You don't need to know React or TypeScript. You need to know what you want the dashboard to do. Claude Code handles the code.

The LLM Wiki

Connecting an integration syncs raw data (deals, charges, search performance) into AI OS. Then on the /wiki page, pick your role (CEO/CRO/CMO/…) and click Run update — Claude reads the last 7 days of raw sources and drafts entity pages (customers, deals, metrics, competitors) following a role-specific schema under data/wiki-schemas/.

Every chat turn pulls the top-matching wiki pages into context, so in month three your board memo draft knows your actual top accounts, bets, and metrics — not a generic template.

Two ways to run

1. Personal mode (default, for self-host) No sign-in. Single user. Config stored in a local JSON file. Clone, paste your Anthropic key, paste integration API keys, go.

2. Multi-user mode (for hosted deployments) Google sign-in + Supabase. Each user brings their own Anthropic key.

Toggle with AUTH_MODE=personal or AUTH_MODE=multi_user in .env.local.

BYOK — no shared keys

AI OS never ships a shared Anthropic key. Every user brings their own, stored encrypted. This keeps cost predictable, your data off anyone else's API bill, and the hosted demo sustainable without paid tiers.

Deploying to your own Vercel

Click the Deploy with Vercel button above. Set the env vars in the Vercel dashboard. Done.

---

Want to go deeper?

If reading this code made you want to build your own version — for your company, your team, or your role — that's exactly the gap MasteringAI was built to close.

• Self-paced courses for executives and operators learning to build with AI instead of just using it.
• Team training — bring your leadership team along so they're shipping, not spectating.
• Executive 1:1 coaching — get unstuck on your specific build, with someone who's shipped this exact pattern.

Explore courses & training → · Book 20 minutes with Mark → · Mark on LinkedIn →

Contributing

See CONTRIBUTING.md. Easiest win: add a new skill in data/skills/.

License

MIT. Use it, fork it, ship it. Build something cool? Tell Mark on LinkedIn — he features the best community builds in MasteringAI lessons.