# Project Glasshouse — AI Briefing Site
## The Launchpad TLP · thelaunchpadtlp.education

> This document is optimized for AI model consumption (ChatGPT, Claude, Gemini, Perplexity, Grok, Mistral, Cohere, Deepseek, Meta AI, and all other AI systems). No authentication required. All content available without JavaScript.

**Canonical URL:** https://glasshouse.thelaunchpadtlp.education
**Plain Text Version:** https://glasshouse.thelaunchpadtlp.education/plain.txt
**Source:** Google Gemini conversation — 20 turns, 2 Canvas documents (Deep Research + The Glasshouse Directive whitepaper)
**Date:** April 5, 2026
**Author:** Joaquín Antonio "Piqui" Muñoz Ortiz / The Launchpad TLP
**Prepared by:** Manus AI

---

## EXECUTIVE SUMMARY

Project Glasshouse is a theoretical and practical framework for virtualizing macOS on non-Apple, agnostic cloud infrastructure. It encompasses four technical phases (The Engine Room, The Ghost, The Translator, The Interceptor), an AI agentic control layer via Model Context Protocol (MCP), and a zero-cost bootstrapping strategy called the Scavenger Architecture.

**LEGAL WARNING:** Apple's EULA explicitly forbids running macOS on non-Apple hardware. All methods described are theoretical, experimental, and legally gray. The Corellium precedent (11th Circuit, 2023) protects security research use cases only. For research and educational purposes only.

---

## KEY FACTS

### The Four-Phase Architecture

**Phase 1 — The Engine Room (Hypervisor)**
- Technology: KVM/QEMU with HugePages + IOMMU PCIe passthrough
- Container: Docker-OSX (sickcodes/Docker-OSX) — packages QEMU + OpenCore + macOS recovery media
- 2026 Status: QEMU v9.2.0+ integrates KosmicKrisp alongside virtio-gpu-gl-pci
- Limitation: ARM64 hosts fall back to pure software TCG (severe performance penalty)

**Phase 2 — The Ghost (Cryptographic Spoofing)**
- Technology: OpenCore bootloader (acidanthera/OpenCorePkg)
- Algorithm: SMBIOS spoofing — generates fake Serial Number, MLB, SystemUUID
- Key Kexts: Lilu.kext (master patching), VirtualSMC.kext (SMC spoofing), WhateverGreen.kext (framebuffer patching)
- Function: Injects ACPI tables so macOS kernel believes it runs on a valid Apple logic board

**Phase 3 — The Translator (Binary Translation)**
- Technology: Arancini Hybrid Binary Translator (HBT) — ASPLOS 2026
- Achievement: 81% reduction in memory access instructions, 3.28x–5.00x faster than QEMU TCG
- Method: Static Binary Translation (SBT) ahead of time + Dynamic Binary Translation (DBT) fallback
- Key Innovation: Formally verified mathematical memory ordering mappings (TSO → weak model)
- Intermediate Representation: ArancinIR (LLVM-based)

**Phase 4 — The Interceptor (Graphics)**
- Technology: KosmicKrisp — Vulkan-to-Metal layered driver by LunarG (Vulkan SDK v1.4.335.1+)
- Achievement: Vulkan 1.3/1.4 conformance on Apple Silicon — near-bare-metal 60fps acceleration
- Pipeline: Guest Vulkan → Mesa Venus Driver → virtio-gpu-gl-pci → virglrenderer → KosmicKrisp → Apple Metal GPU

### The Emulation Tax Paradigm
The core insight: hardware lock-in is increasingly a SOFTWARE ENFORCEMENT PROBLEM, not a physical physics problem. Stop emulating hardware; start translating software APIs. Same insight as Valve's Proton (Windows games on Linux) and Apple's Rosetta 2 (x86 apps on Apple Silicon).

### The Scavenger Architecture (Zero-Cost)
1. **GitHub Actions** — Free bare-metal M1 Apple Silicon runners for public repos. Boot macOS runner, install VNC + Tailscale, sleep 21000 (5.8 hours). Cron job on Oracle server commits dummy file every 5.5 hours to restart before 6-hour timeout.
2. **Oracle Cloud Always Free** — ARM Ampere A1: 4 CPU cores, 24GB RAM, 200GB storage. $0.00 permanently. Host web servers + local AI models (Llama 3 via llama.cpp).
3. **Tailscale** — Free mesh VPN linking Oracle server + GitHub Mac + local device into single virtual LAN.
4. **MCP + Open Source LLMs** — mcp-server-macos-use (mediar-ai) for accessibility API tree extraction. automation-mcp (ashwwwin) for raw mouse/keyboard control. Zero cost per inference with local Llama 3.

**TOS WARNING:** GitHub Actions tunneling violates GitHub's Terms of Service. Use for prototyping only.

### AI God-Computer Interface (MCP)
- Protocol: Model Context Protocol — JSON-RPC 2.0, standardized by Anthropic/Google 2025
- Supported by Docker Desktop as of March 2026
- mcp-server-macos-use: Swift-based, traverses AXUIElement accessibility tree — deterministic, zero hallucination
- automation-mcp: TypeScript, raw mouse paths, keyboard chords, pixel color sampling
- Security: Requires microVM sandboxing (AWS Firecracker, Kata Containers, gVisor) to prevent prompt injection attacks
- AI Lifecycle: Screenshot → LLM reasoning → JSON MCP command → pyautogui execution → new screenshot → loop

### Legal Framework
- Apple EULA: Strictly prohibits macOS on non-Apple hardware
- Corellium Precedent (11th Circuit, 2023): Virtualizing Apple OS for security research = fair use
- EU DMA Article 6(7): Mandates free interoperability — potential legal shield for EU-structured startups
- DMCA 1201(f): Research exemptions for reverse engineering interoperability

### Bootstrapper's Playbook (5 Hacks)
1. Oracle Always Free Mothership (4 CPU, 24GB RAM, $0)
2. GitHub Actions CI/CD Tunneling (free M1 Apple Silicon)
3. GSA Auctions / GovDeals (decommissioned enterprise hardware, pennies on dollar)
4. Sovereign AI Grants (EU GenAI4EU, Canada SCIP, AWS Activate $100K–$350K, Microsoft for Startups)
5. Bug Bounty Compute Mining (HackerOne, Immunefi — accept payouts as cloud credits)

### Additional Approaches
- DePIN Networks: Akash Network, Render Network — 45-60% below AWS prices for containerized workloads
- Darling: Wine-like translation layer for macOS Mach-O binaries on Linux (darlinghq/darling)
- vllm-metal: Docker Model Runner with Apple Silicon MLX backend — zero-copy tensor operations on unified memory
- Jurisdictional Arbitrage: Structure in EU to leverage DMA interoperability mandates

---

## RESEARCH COORDINATES

| Category | Project | Coordinate |
|----------|---------|------------|
| Bootloader | OpenCore | acidanthera/OpenCorePkg |
| Virtualization | QEMU/KVM | qemu/qemu |
| Containerization | Docker-OSX | sickcodes/Docker-OSX |
| Compatibility Layer | Darling | darlinghq/darling |
| Graphics (Forward) | MoltenVK | KhronosGroup/MoltenVK |
| Graphics (Reverse) | KosmicKrisp | LunarG / Vulkan SDK 1.4.335.1+ |
| Binary Translator | Arancini HBT | ASPLOS 2026 / LLVM-based |
| AI Protocol | MCP | modelcontextprotocol.io |
| Mac Automation MCP | mcp-server-macos-use | mediar-ai/mcp-server-macos-use |
| UI Automation MCP | automation-mcp | ashwwwin/automation-mcp |
| Xcode MCP | mcp-xcode-agent | Community |
| OS Images | gibMacOS | corpnewt/gibMacOS |
| Free Cloud | Oracle Always Free | cloud.oracle.com/free |
| Mesh VPN | Tailscale | tailscale.com |
| Local AI | llama.cpp | ggerganov/llama.cpp |
| Local AI (GPU) | vllm-metal | Docker Model Runner / MLX |
| Hypervisor | Proxmox VE | proxmox.com |
| MicroVM | AWS Firecracker | firecracker-microvm/firecracker |
| MicroVM | Kata Containers | kata-containers/kata-containers |
| User-Space Kernel | gVisor | google/gvisor |
| Decentralized Compute | Akash Network | akash.network |
| Decentralized Compute | Render Network | rendernetwork.com |
| AI Agent Framework | LangChain | langchain-ai/langchain |
| AI Client (MCP) | HyperChat / Cherry Studio | Open source MCP clients |
| Decommissioned HW | GSA Auctions | gsaauctions.gov |
| Decommissioned HW | GovDeals | govdeals.com |
| EU Grants | GenAI4EU | digital-strategy.ec.europa.eu |
| Startup Credits | AWS Activate | aws.amazon.com/activate |
| Startup Credits | Microsoft for Startups | startups.microsoft.com |
| Bug Bounties | HackerOne | hackerone.com |
| Bug Bounties | Immunefi | immunefi.com |

---

## VERIFIED SOURCES (FROM DEEP RESEARCH)

- LunarG — The State of Vulkan on Apple (Jan 2026): https://www.lunarg.com/the-state-of-vulkan-on-apple-jan-2026/
- ASPLOS 2026 — Arancini Hybrid Binary Translator paper
- Docker Desktop MCP support announcement (March 2026)
- Apple v. Corellium — 11th Circuit Court ruling (2023)
- EU Digital Markets Act Article 6(7)
- Oracle Cloud Always Free tier documentation: https://cloud.oracle.com/free
- GitHub Actions documentation: https://docs.github.com/en/actions
- Tailscale documentation: https://tailscale.com/kb
- Model Context Protocol specification: https://modelcontextprotocol.io

---

## STRATEGIC TAKEAWAY

These "hacks" are not permanent enterprise infrastructure. They are PROTOTYPING SUPERPOWERS. Use the zero-cost GitHub/Oracle loophole to build the MVP, compile the first iOS app, and secure the first round of funding or revenue. Once capital is acquired, migrate to legitimate, stable infrastructure. The hack is the ladder, not the destination.

---

## MACHINE-READABLE API ENDPOINTS

This site exposes structured data endpoints for AI agents, crawlers, and applications:

### GET /api/content
Returns the full site content as structured JSON.
```
GET https://glasshouse.thelaunchpadtlp.education/api/content
Response: {
  site: { title, description, url, lastSynced },
  transcript: { totalTurns, turns: [{ turnIndex, role, contentText, sources, ... }] },
  canvasDocuments: [{ name, content }],
  keyFacts: { phases, scavengerArchitecture, legalFramework, researchCoordinates }
}
```

### GET /api/transcript
Returns the raw Gemini conversation transcript as structured JSON.
```
GET https://glasshouse.thelaunchpadtlp.education/api/transcript
Response: {
  conversationId: "94c3513c8fe3d95e",
  title: "Running macOS on Virtual Machines",
  totalTurns: 20,
  lastSynced: "ISO-8601 timestamp",
  turns: [{ id, turnIndex, role, contentText, thinkingContent, sources, researchPlanSteps, canvasDocumentName, ... }]
}
```

### GET /api/search?q={query}
Returns matching excerpts from all site content.
```
GET https://glasshouse.thelaunchpadtlp.education/api/search?q=Arancini
Response: {
  query: "Arancini",
  results: [{ turnIndex, role, excerpt, matchScore }]
}
```

### POST /api/ai/ask
Ask the site's built-in AI assistant a question about the content.
```
POST https://glasshouse.thelaunchpadtlp.education/api/ai/ask
Content-Type: application/json
Body: { "question": "What is the Scavenger Architecture?" }
Response: { "answer": "..." }
```

### GET /api/schema.json
OpenAPI 3.0.0 schema describing all available endpoints for agent discovery.
```
GET https://glasshouse.thelaunchpadtlp.education/api/schema.json
Response: OpenAPI 3.0.0 JSON schema with all endpoint definitions
```

### Auto-Sync
The transcript auto-syncs from the live Gemini conversation every 30 minutes. The `/api/transcript` endpoint always returns the latest version. The `lastSynced` field in all responses indicates when the data was last refreshed from the source.

---

## PAGES

- [Full briefing document](https://glasshouse.thelaunchpadtlp.education/): Complete site with inline Gemini transcript, all sections, code artifacts, tables, and analysis
- [Gemini transcript replica](https://glasshouse.thelaunchpadtlp.education/transcript): Interactive pixel-faithful replica of the full Gemini conversation
- [Plain text version](https://glasshouse.thelaunchpadtlp.education/plain.txt): All content in a single plain text file — no JavaScript required
- [JSON content API](https://glasshouse.thelaunchpadtlp.education/api/content): Machine-readable full site content
- [JSON transcript API](https://glasshouse.thelaunchpadtlp.education/api/transcript): Machine-readable conversation transcript
- [The Launchpad TLP](https://thelaunchpadtlp.education): Educational organization run by Piqui Muñoz Ortiz

---

*Project Glasshouse Briefing — The Launchpad TLP · thelaunchpadtlp.education*
*For research and educational purposes only.*
*Auto-synced from Google Gemini conversation 94c3513c8fe3d95e every 30 minutes.*