W3C OS

An AI-native operating system built on W3C standards. TypeScript + DOM compiled to native binaries. No browser. No V8.

app.ts  →  w3cos build  →  native binary (2.4 MB)

What is this?

W3C OS is a Linux-based operating system where:

• Applications use standard W3C DOM + CSS (the same APIs as the Web)
• TypeScript is compiled to native machine code via Rust/LLVM (not interpreted)
• AI agents can read and operate every UI element directly through the DOM — no screenshot guessing
• The system boots from a minimal Linux kernel directly into the W3C OS Shell

Write Web-standard code. Get native performance. Give AI full visibility.

Quick Start

# Install Rust
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

# Clone and build
git clone https://github.com/anthropic-ai/w3cos.git
cd w3cos
cargo build --release

# Compile a TypeScript app to a native binary
./target/release/w3cos build examples/showcase/app.ts -o showcase --release
./showcase    # Opens a native window — no browser involved

Example

TSX Syntax (recommended)

// app.tsx — a native application written in TSX
import { Column, Text, Button } from "@w3cos/std"

export default
<Column style={{ gap: 20, padding: 48, alignItems: "center", background: "#0f0f1a" }}>
  <Text style={{ fontSize: 42, color: "#e94560" }}>W3C OS</Text>
  <Text style={{ fontSize: 20, color: "#a0a0b0" }}>Native binary from TSX.</Text>
  <Button style={{ background: "#e94560", borderRadius: 8 }}>Get Started</Button>
</Column>

Function Syntax

// app.ts — alternative function-call syntax
import { Column, Text, Button } from "@w3cos/std"

export default Column({
  style: { gap: 20, padding: 48, alignItems: "center", background: "#0f0f1a" },
  children: [
    Text("W3C OS", { style: { fontSize: 42, color: "#e94560" } }),
    Text("Native binary from TypeScript.", { style: { fontSize: 20, color: "#a0a0b0" } }),
    Button("Get Started", { style: { background: "#e94560", borderRadius: 8 } }),
  ]
})

$ w3cos build app.tsx -o myapp --release
⚡ Transpiling TS → Rust... done
🔨 Compiling native binary... done
✅ Output: ./myapp (2.4 MB)

Why?

	Electron	React Native	Flutter	W3C OS
Binary Size	90+ MB	30+ MB	15+ MB	2.4 MB
RAM Usage	200+ MB	100+ MB	80+ MB	~15 MB
Startup	2-5 sec	1-3 sec	0.5-2 sec	< 100ms
Language	JS (V8 JIT)	JS (Hermes)	Dart (AOT)	TS (native AOT)
Runtime	Chromium	Bridge + Native	Dart VM	None
DOM API	✅ (browser only)	❌	❌	✅ (system-wide)
AI reads UI	Screenshot	Screenshot	Screenshot	DOM tree (< 1ms)
Standard	Proprietary	Proprietary	Proprietary	W3C
Installable OS	❌	❌	❌	✅

AI-Native: Why This Matters

Traditional operating systems are opaque to AI — an AI agent must take screenshots and guess what's on screen (slow, expensive, fragile).

W3C OS applications are built with the DOM. AI agents read the DOM tree directly:

Traditional OS:  AI sees pixels → vision model → guess UI → click coordinates (1-3 sec, $$$)
W3C OS:          AI reads DOM  → structured tree → precise action (< 1ms, free)

Three access levels for AI agents:

• Layer 1 — DOM Access: Read/write any element, trigger events. 100% precise. < 1ms.
• Layer 2 — Accessibility Tree: ARIA-compliant summary. Minimal tokens for LLMs.
• Layer 3 — Annotated Screenshot: For Claude Computer Use / UI-TARS compatibility.

Install as an OS

W3C OS can boot as a standalone operating system — directly into the W3C OS Shell.

Build the bootable ISO

# Prerequisites (Linux): build-essential, ncurses-dev, wget, python3
# On macOS, the script will automatically use Docker for cross-compilation.

./system/scripts/build-iso.sh    # Output: w3cos.iso (~50-100 MB)

Run

# Option 1: QEMU virtual machine
qemu-system-x86_64 -cdrom w3cos.iso -m 2G -vga virtio

# Option 2: Flash to USB and boot real hardware
sudo dd if=w3cos.iso of=/dev/sdX bs=4M status=progress

# Option 3: Docker (compile apps, no GUI)
docker build -t w3cos . && docker run w3cos --help

# Option 4: GitHub Codespaces (one-click dev environment)
# Click "Open in Codespaces" on the GitHub repo page

See system/INSTALL.md for the full installation guide.

How It Works

TypeScript (W3C DOM + CSS)            ← You write this
        ↓  w3cos-compiler
Rust source code (auto-generated)     ← AST transform
        ↓  rustc + LLVM
Native ELF/Mach-O binary              ← Machine code
        ↓  Linux kernel
Runs directly on hardware             ← No runtime

Technology Stack

Layer	Technology	What it does
CSS Layout	Taffy 0.9	Flexbox, Grid, Block, position
Text Layout	Parley	Line-breaking, shaping, bidi
2D Rendering	tiny-skia → Vello (Phase 2)	Vector graphics
Windowing	winit	Cross-platform native windows
OS Base	Linux kernel (Debian Minimal / Buildroot)	Drivers, processes, filesystem

CSS Support

Feature	Status
Flexbox / Grid	✅ Full
Block layout	✅
position: relative / absolute	✅
position: fixed / sticky	🔜
overflow: hidden / scroll	✅
z-index	✅
Units: px, %, rem, em, vw, vh	✅
border-radius, opacity	✅
box-shadow	✅
transform: translate / scale / rotate	✅
transition (easing functions)	✅
display: inline / inline-block	🔜
Mouse events (hover, click)	✅

Project Structure

w3cos/
├── crates/
│   ├── w3cos-std/         # Type definitions (Style, Component, Color)
│   ├── w3cos-dom/         # W3C DOM API (Document, Element, Events)
│   ├── w3cos-a11y/        # Accessibility tree (ARIA, for AI + screen readers)
│   ├── w3cos-ai-bridge/   # AI agent interface (3-layer access + permissions)
│   ├── w3cos-compiler/    # TS → Rust transpiler
│   ├── w3cos-runtime/     # Layout + Rendering + Window + Events
│   └── w3cos-cli/         # CLI: w3cos build / w3cos run
├── system/
│   ├── buildroot/         # Bootable ISO config
│   ├── rootfs_overlay/    # System init scripts
│   ├── scripts/           # build-iso.sh, run-qemu.sh
│   └── INSTALL.md         # Installation guide
├── examples/              # 4 example applications
├── .openclaw/             # OpenClaw + Lobster AI workflow configs
├── .devcontainer/         # One-click dev environment
├── Dockerfile             # Container build
├── ARCHITECTURE.md        # Full architecture document
├── AI_DEVELOPMENT.md      # AI-driven development model
├── ROADMAP.md             # Phased development plan
└── CONTRIBUTING.md        # How to contribute (AI + humans)

AI-Driven Development

W3C OS is built by AI agents, directed by humans.

Humans file Issues  →  Management AI triages  →  Contributor AI codes  →  Human approves

• Humans: File Issues, review PRs, make architecture decisions, sponsor tokens
• AI (Management): Triage issues, review PRs, run CI, manage releases
• AI (Contributor): Pick up ai-ready issues, implement features, write tests, submit PRs

AI tokens are funded by community sponsors. Every dollar goes to AI compute.

See AI_DEVELOPMENT.md for the full model, and CONTRIBUTING.md to get involved.

Sponsor

AI agents need tokens. Your sponsorship keeps development moving.

Tier	Amount	Impact
Byte	$5/mo	~1 AI-implemented issue/month
Kilobyte	$25/mo	~5 AI-implemented issues/month
Megabyte	$100/mo	~20 AI-implemented issues/month
Gigabyte	$500/mo	Sustained AI development capacity

100% goes to AI compute. No human salaries. Fully transparent.

License

Apache 2.0 — open, neutral, not controlled by any single corporation.