CortexOS

A concept demo built around one central idea: the AI assistant isn't a bolted-on app sitting alongside everything else — it's positioned as the operating system's control layer, with the standing ability to rearrange the entire desktop, window layout, and system theme from a single typed or spoken instruction, instead of you dragging and clicking everything into place yourself.

It ships as two fully interactive, self-contained desktop simulations — a Windows-style environment and a macOS-style environment — both built around the same underlying idea: a persistent AI control layer that can reorganize the whole workspace on command.

This is a front-end demo, not a real operating system or a real AI. Every "app" runs entirely in your browser tab, all data is stored in localStorage, and the AI control layer is a scripted sequence of DOM changes — there is no LLM call, no network request, and no real speech recognition anywhere in this project. See How "AI" actually works here for the honest breakdown.

Quick start

There is no build step, no server, and no dependencies to install.

1. Open index.html in a modern desktop browser (Chrome, Edge, or Firefox recommended).
2. Pick Windows Environment or macOS Environment.
3. Press Ctrl+Space (or click the sparkle icon) to open the AI control layer.

That's it — everything is client-side. You can also open windows.html or macos.html directly to skip the launcher.

What's actually in the box

CortexOS/
├── index.html     # launcher — choose Windows or macOS
├── windows.html   # Windows-style desktop simulation
├── macos.html     # macOS-style desktop simulation
└── README.md

Each of the three pages is a single self-contained HTML file (Tailwind CSS via CDN + vanilla JavaScript, no framework, no bundler). windows.html and macos.html are independent — they don't share a script file — but they're built from the same architecture, so if you understand one, you understand the other. See the comments at the top of each <script> block for an orientation.

The AI control layer

Both environments have an identical AI layer, just re-skinned for the OS it lives in:

	Windows	macOS
Open it	Ctrl+Space, or the sparkle icon on the taskbar	Ctrl+Space, or the sparkle icon in the menu bar (where Siri would live)
Talk to it	Type a sentence, or click the mic for simulated voice input	Same

You can type things like:

• "enter focus mode" / "switch to creative mode" / "let's relax" — triggers one of the three preset workspace transformations below
• "open notepad" / "close the browser" — opens or closes a specific app by name
• Anything else gets a graceful "here's what I can actually do" fallback instead of silently failing

The three workspace modes

Each mode is a scripted, narrated sequence — the AI layer logs each step it's "taking" (e.g. "Closing entertainment apps…", "Snapping windows side-by-side…") with a short pause between steps so the reorganization is visible rather than instant.

• Focus — closes distraction apps (browser, media player), opens the Code Editor and Terminal snapped side-by-side, dims non-essential desktop elements, and shifts the whole environment to a dark, distraction-free blue theme.
• Creative — closes unrelated apps, opens Design Studio in a wide layout with a small reference-file panel, and shifts to a warm, high-contrast theme.
• Relax — closes productivity apps, opens and centers the Music/Media player, hides taskbar/Dock running-app badges, and shifts to a soft warm evening theme.
• Reset — reverts the theme, dimming, and badge changes without closing anything.

Simulated voice input

Click the microphone icon to enter a "listening" state. Instead of using a real microphone, it shows a list of example phrases grouped by intent (rearrange the workspace / open an app / close an app). Clicking one simulates a transcription typing itself into the input field, then runs it through the exact same command parser as typed text.

Explainer bubbles

Hovering the AI trigger, the mic button, each mode preset, or any desktop icon shows a small tooltip explaining what that feature does — and, for apps specifically, whether it's a fully working simulation or a static placeholder (see the table below).

What's actually functional vs. a placeholder

This project mixes real, working mini-apps with static mockups. Nothing is hidden — every desktop icon has a hover bubble stating which category it falls into, and here's the same information in one place:

App	Windows name	macOS name	Status
File browser	This PC	Finder	Placeholder — static drive/folder tiles, not a real file browser
Trash	Recycle Bin	Trash	Placeholder — always shows empty, no real delete/restore
Text editor	Notepad	TextEdit	Functional — type, then File → Save/Open persists text to localStorage
Browser	Microsoft Edge	Safari	Partly functional — address bar, mock search, and back/forward history work; results are fabricated, there's no real internet access
Settings	Settings	System Settings	Functional — page navigation and the Dark Mode toggle really work; device/network/Bluetooth info is static
Terminal	Command Prompt	Terminal	Functional mini shell — supports help, dir, date, time, cls, echo
Code Editor	Code Editor	Code Editor	Functional text editor with live line numbers (not persisted between sessions)
Design Studio	Design Studio	Design Studio	Functional — real freehand drawing on an HTML canvas with brush/eraser/color tools
Media player	Media Player	Music	Simulated — play/pause and a progress bar run on a timer; there's no real audio file

How "AI" actually works here

To be explicit about what's real and what isn't:

• No model, no API calls. detectMode() and findAppInText() are plain regex/substring matching against a small keyword list — not natural language understanding.
• No speech recognition. The mic button never requests microphone access. It reveals a list of pre-written example phrases; picking one plays a typing animation and then runs the same text-command parser used for keyboard input.
• The "reorganization" is a fixed script. Each mode function (runFocusMode, runCreativeMode, runRelaxMode) is a hand-written sequence of window-manager calls with await sleep(ms) between them, purely so a human can watch each step happen.

The goal of this project is to make the interaction pattern of an AI-driven OS — the AI positioned as the control layer, not a bolted-on app — tangible and demoable, not to actually implement one.

Window manager features (both environments)

• Drag windows by their title bar; text on the desktop and inside windows never gets accidentally selected mid-drag.
• Windows can't be dragged fully off-screen — at least part of the title bar always stays reachable.
• Resize from the bottom-right corner.
• Minimize, maximize/restore, and close, each with a short eased animation.
• Drag a window to the left/right screen edge to snap it to half-width; drag it away to restore its original size. Drag to the top edge to maximize.
• Click a taskbar/Dock icon to toggle its window: opens it if not running, restores+focuses it if minimized, minimizes it if already focused, or focuses it if running but unfocused.
• Right-click the desktop for a themed context menu (Refresh/Clean Up flashes the desktop, "New Text Document/File" adds a real, savable file to the desktop).
• A working light/dark theme toggle (Settings → Personalization on Windows, Settings → Appearance on macOS) that also feeds the AI control layer's per-mode accent colors.

Tech stack

• Plain HTML/CSS/JS — no build tooling
• Tailwind CSS via the Play CDN, for utility layout only
• Inline SVG for every icon (no icon fonts, no image assets, no external requests besides the Tailwind CDN script)
• localStorage for the only "persistence" that exists (theme, saved text files, desktop icons)

Known limitations

• Single window instance per app — opening an already-running app focuses/restores it rather than opening a second copy.
• No real file system, no real network access, no real audio or speech.
• Not tested for mobile/touch input — this is a desktop-pointer (mouse) experience.
• The two environments are independent files with duplicated logic by design (each is meant to be readable end-to-end on its own), so a fix in one does not automatically apply to the other.

License

No license has been added yet — treat this as "all rights reserved" until a LICENSE file is added.