AI-Native · Privacy-First · Open Source

The Operating System that generates its own apps

No pre-built apps. No binaries. Describe what you need in plain language, and the operating system generates it, sandboxes it, and runs it. Inspired by Karpathy's LLM OS concept.

View on GitHub Get Started Download VM
LLM OS v0.1.0
> make me a pomodoro timer with break reminders [gateway] Routing to Ollama (qwen2.5:14b) — simple complexity [gateway] Generated in 3.2s [analyzer] Static analysis: 0 critical, 0 warnings [caps] App requests: [ui:window, timer:basic, storage:local] [user] Approved capabilities [sandbox] Launched app-1 in isolated iframe [sandbox] App running with 3 capabilities granted

Core Values

These aren't guidelines. They're enforced at every layer — deterministic scans, AI review, and human oversight.

01

Protect the user first

No telemetry. No tracking. No data exfiltration. Generated apps run in sandboxes with strict capability gates. When in doubt, deny access. User privacy is non-negotiable.

02

Empower the user

No artificial limits. No paywalls. Users can generate and run any software they want — as long as it doesn't harm others. The OS serves the user, not the other way around.

03

Take a piece, leave a piece

Use it freely. Adapt it as you see fit. But if you benefit from it — contribute back, even a little. Code that's contributed must not damage the core idea.

04

Nothing is perfect

These rules aren't perfect. Neither is this code. We can always improve — as long as the core intent isn't violated. Ship working code, iterate, improve.

How It Works

From natural language to a running app in seconds. Every step is security-gated.

1

You describe what you need

"Make me a todo list with categories" — plain language, no code required.

2

The kernel generates the app

Routes to the best LLM — local Ollama for simple apps, Claude API for complex ones. Prompt injection is stripped before generation.

3

Static analysis scans the code

Deterministic regex/AST scan blocks eval(), dynamic imports, parent frame access, and encoded payloads. No LLM in the loop — no recursive injection.

4

You approve the capabilities

The app declares what it needs (storage, timers, network). You review and approve each one. The app gets nothing you don't explicitly allow.

5

It runs in a sandbox

Isolated iframe with strict CSP. The SDK communicates with the kernel via postMessage. Every call is validated against your approved capabilities.

Architecture

Layered design — every layer gets replaced as we move toward a custom kernel.

Phase 1 (current) Phase 4 (goal) Generated Apps Generated Apps iframe sandbox WASM sandbox | | Shell UI Shell UI browser DOM framebuffer | | Kernel Kernel Node.js Rust no_std | | Host OS Custom Microkernel Windows/Linux/macOS UEFI boot, zero Linux

Get Started

Three paths. Pick the one that fits.

RECOMMENDED

With Claude Code

Open the repo in VS Code. Claude Code reads CLAUDE.md automatically and knows the entire project.

gh repo fork DayZAnder/llm-os --clone
cd llm-os
code .

# Tell Claude Code what to build:
# "Add WASM sandbox"
# "Improve the shell UI"
# "Audit the security"
ANY AI TOOL

With ChatGPT, Copilot, etc.

Copy a ready-made prompt from CONTRIBUTING.md into your preferred AI assistant.

gh repo fork DayZAnder/llm-os --clone
cd llm-os

# Open CONTRIBUTING.md
# Copy a component prompt
# Paste into your AI tool
# Each prompt includes values context
NO AI NEEDED

Just code

Fork, read the README, run the prototype, and pick an issue.

gh repo fork DayZAnder/llm-os --clone
cd llm-os
cp .env.example .env
node src/server.js

# Open http://localhost:3000
# Pick an issue from GitHub

Values Enforcement

Three layers, no single point of trust. Every contribution is checked.

1

Deterministic Scan

Regex-based static analysis runs locally and in CI. Detects telemetry, sandbox weakening, privacy violations, tracking code. Blocks merge on critical findings.

2

AI Values Guardian

Claude reviews every PR diff against the core values. Posts findings as comments. Catches subtle violations that regex can't see.

3

Human Review

PR template requires values self-certification. Maintainer has final authority on edge cases. No automated system is trusted alone.

Download LLM OS

Boot a full LLM OS instance in your hypervisor. Alpine Linux + Docker + Node.js — ready to generate apps.

Q

QCOW2

~389 MB (compressed)

For Proxmox, KVM, QEMU, and libvirt. Import as a VM disk image.

Download QCOW2
V

VHDX

~904 MB

For Hyper-V on Windows. Create a Gen 1 VM and attach as the primary disk.

Download VHDX
Quick start: Boot the VM, log in as root / llmos, then open http://<vm-ip>:3000 in your browser. Configure your LLM backend with llmos-config set OLLAMA_URL http://your-ollama:11434. Change the default password on first login!

View release notes on GitHub

This OS is built with AI,
by AI, for humans

The next operating system won't ship apps — it'll generate them.
If that future interests you, start building.

Contribute on GitHub Read CONTRIBUTING.md

Quick links: llm-os.dev/#start · llm-os.dev/#contribute