================================================================================
PROJECT GLASSHOUSE — COMPLETE BRIEFING DOCUMENT
The Launchpad TLP · thelaunchpadtlp.education
Prepared by Manus AI for Joaquín Antonio "Piqui" Muñoz Ortiz
April 5, 2026
================================================================================

Source: Google Gemini conversation — 20 turns, 2 Canvas documents
Canvas 1: "Deep Research and Fact-Checking" (31,277 chars) — Completed April 5, 2026
Canvas 2: "The Glasshouse Directive" whitepaper — God Mac.pdf

LEGAL NOTICE: 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.

AI MANIFEST:  https://glasshouse.thelaunchpadtlp.education/llms.txt
THIS FILE:    https://glasshouse.thelaunchpadtlp.education/plain.txt
FULL SITE:    https://glasshouse.thelaunchpadtlp.education/

================================================================================
SECTION 00: EXECUTIVE SUMMARY
================================================================================

Project Glasshouse is a theoretical and practical framework for virtualizing macOS
on non-Apple, agnostic cloud infrastructure. The computing landscape is undergoing
a radical shift away from x86 architecture toward ARM-based processing, typified by
Apple's transition to Apple Silicon.

This briefing analyzes the state-of-the-art technologies that have shattered the
"Emulation Tax," including hybrid binary translators (Arancini) and Vulkan-to-Metal
graphics pipelines (KosmicKrisp). It explores how modern AI agents securely host and
operate these operating systems using microVM sandboxing and the Model Context
Protocol (MCP). Finally, it outlines a radical, asymmetrical strategy for bootstrapped
startups to achieve hyperscale macOS computing with zero capital.

KEY METRICS:
  Core Breakthrough: KosmicKrisp + Arancini eliminate the Emulation Tax
                     5x faster, 81% fewer memory ops vs QEMU TCG
  AI Interface:      MCP + mcp-server-macos-use gives AI agents
                     deterministic, hallucination-free macOS control
  Zero-Cost Path:    GitHub Actions (free M1) + Oracle Always Free (24GB ARM)
                     + Tailscale = $0.00 infrastructure

================================================================================
SECTION 01: THE ARGUMENT IN PLAIN LANGUAGE
================================================================================

The core claim of this conversation is that it is theoretically and practically
possible to run Apple's macOS on non-Apple hardware — ranging from generic cloud
servers to free CI/CD runners — and to fully automate these environments using AI
agents. The conversation moves from traditional "Hackintosh" methods to a highly
advanced theoretical architecture called Project Glasshouse.

The iOS App Store is one of the most lucrative digital markets in the world, but
the entry ticket is a $1,000+ Apple computer. For developers and researchers in the
Global South, or with zero capital, this represents a structural barrier to economic
participation. Project Glasshouse proposes eliminating this barrier entirely.

With Apple planning to release macOS 27 exclusively for Apple Silicon, and macOS 28
slated to remove Rosetta 2 entirely by 2027, the traditional "Hackintosh" methodology
has reached its terminal end of life. Hypervisor-based virtualization and advanced
hardware emulation have become the sole viable pathways for executing Apple operating
systems on non-Apple host infrastructure.

KEY QUOTE: "To build a 'God-Mac' cloud infrastructure with absolutely zero capital,
we have to abandon the traditional path of renting dedicated servers. When you have
no money, your primary currency is cunning, open-source leverage, and exploiting
corporate free tiers to their absolute limits."
  — Google Gemini, Project Glasshouse Conversation, April 2026

================================================================================
SECTION 02: THE FOUR-PHASE MASTER ARCHITECTURE
================================================================================

Project Glasshouse is a four-phase software stack designed to detach Apple's
software from Apple's hardware.

PHASE 1 — THE ENGINE ROOM (Hypervisor & Host OS)
  Technology:   KVM/QEMU with HugePages memory management + 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)

  Host Provisioning Script:
    apt-get update -y && apt-get install -y qemu-kvm libvirt-daemon-system \
      libvirt-clients bridge-utils ovmf
    systemctl enable --now libvirtd
    sed -i 's/GRUB_CMDLINE_LINUX_DEFAULT="/GRUB_CMDLINE_LINUX_DEFAULT="amd_iommu=on \
      iommu=pt kvm.ignore_msrs=1 /g' /etc/default/grub
    update-grub && update-initramfs -u
    echo "vm.nr_hugepages = 1024" >> /etc/sysctl.conf && sysctl -p

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 engine — hooks into XNU kernel during boot
    VirtualSMC.kext:   Emulates Apple's System Management Controller (SMC)
    WhateverGreen.kext: Patches macOS framebuffers to prevent "Black Screen" on boot
  Function:     Injects ACPI tables so macOS kernel believes it runs on a valid
                Apple logic board

PHASE 3 — THE TRANSLATOR (Binary Translation — Arancini HBT)
  Technology:   Arancini Hybrid Binary Translator (HBT) — ASPLOS 2026
  IR:           ArancinIR (LLVM-based)
  Method:       Static Binary Translation (SBT) ahead of time
                + Dynamic Binary Translation (DBT) fallback
  Key Innovation: Formally verified mathematical memory ordering mappings
                  (TSO → weak model)
  Performance vs QEMU TCG:
    Memory access instructions: 19% of baseline (81% reduction)
    Execution speed:            3.28x to 5.00x faster (geometric mean)
    Memory model safety:        Formally verified (vs heuristic barrier injection)
    Multi-thread correctness:   Provably correct (vs best effort)

PHASE 4 — THE INTERCEPTOR (Graphics — KosmicKrisp)
  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:
    Stage 1: Guest Application — issues standard Vulkan API rendering commands
    Stage 2: Mesa 3D (Venus Driver) — intercepts Vulkan calls, packages SPIR-V shaders
    Stage 3: virtio-gpu-gl-pci — paravirtualized kernel device transfers data
    Stage 4: virglrenderer — receives memory pages, reconstructs Vulkan API calls
    Stage 5: KosmicKrisp — translates Vulkan commands natively into Apple Metal
    Stage 6: Apple Silicon GPU — executes Metal instructions natively

================================================================================
SECTION 03: TECHNICAL COMPENDIUM
================================================================================

I. CORE VIRTUALIZATION & HYPERVISOR STACK

KVM/QEMU:
  Coordinates:  github.com/qemu/qemu
  Logic:        Uses KVM (Kernel-based Virtual Machine) for near-native CPU execution
  Requirement:  Intel VT-x or AMD-V + AVX2 instruction sets
  Integration:  config.plist must be tuned for Q35 chipset
  Standard:     virtio-gpu-pci for high-speed I/O

Docker-OSX:
  Coordinates:  github.com/sickcodes/Docker-OSX
  Architecture: Wraps QEMU inside a Docker container — macOS-as-a-Service
  2026 Status:  Supports macOS 26 (Tahoe) images out-of-the-box
  Use Case:     Ideal for headless CI/CD and automated Xcode builds

Apple Virtualization.framework (Native Only):
  High-level API for creating VMs on native Apple hardware only
  Requires VZVirtualMachineConfiguration + VZMacPlatformConfiguration
  Uses VZMacOSInstaller to extract and boot from .ipsw restore images
  Employs com.apple.security.virtualization entitlements — cannot be ported to Linux

II. HARDWARE SPOOFING & BOOTLOADERS

OpenCore (OC):
  Coordinates:  github.com/acidanthera/OpenCorePkg
  Algorithm:    SMBIOS spoofing — generates fake Serial Number, MLB, SystemUUID
  Design Kit:   OCAuxiliaryTools (OCAT) for visual configuration of config.plist

Key Kernel Extensions:
  Lilu.kext:         Master patching engine — hooks into XNU kernel during boot,
                     injects arbitrary code into protected memory
  VirtualSMC.kext:   Emulates Apple's System Management Controller (SMC)
                     Intercepts OS calls and returns mathematically valid spoofed responses
  WhateverGreen.kext: Patches macOS framebuffers to force standard VESA/DisplayPort outputs

III. API TRANSLATION & GRAPHICS

KosmicKrisp (New in 2025/2026):
  Source:       LunarG / Vulkan SDK v1.4.335.1+
  Technology:   Layered driver intercepting and translating Vulkan API commands
                directly into Metal API commands with minimal translation overhead
  Achievement:  Vulkan 1.3/1.4 conformance on Apple Silicon
  Pipeline:     Sits between virglrenderer (deserialization) and Apple Silicon GPU
  Reference:    https://www.lunarg.com/the-state-of-vulkan-on-apple-jan-2026/

MoltenVK (Complementary):
  Coordinates:  github.com/KhronosGroup/MoltenVK
  Direction:    Runs Vulkan on Metal (forward direction)
  Provides:     Mathematical logic for SPIR-V to AIR (Apple Intermediate Representation)

IV. BINARY TRANSLATION — ARANCINI FRAMEWORK (ASPLOS 2026)

The x86_64 architecture operates on Total Store Ordering (TSO) — a strong memory
model. ARM and RISC-V use weak memory models. Traditional DBTs inject heavy memory
barriers between every translated operation, crippling performance.

Arancini is a Hybrid Binary Translator (HBT) using LLVM and ArancinIR. It performs
Static Binary Translation (SBT) ahead of time, then seamlessly falls back to Dynamic
Binary Translation (DBT) for dynamic control flows. Formally verified mathematical
mapping schemes guarantee strong memory semantics on ARM and RISC-V without
brute-force barriers.

Benchmark Results (Phoenix + EEMBC suites):
  Memory Access Instructions: 81% reduction vs QEMU TCG
  Execution Speed:            3.28x to 5.00x faster
  Memory Model Safety:        Formally verified
  Multi-thread Correctness:   Provably correct

V. LOCAL AI INFERENCE — vLLM-METAL ARCHITECTURE

Docker Model Runner (DMR) integrates the vllm-metal backend, routing inference calls
from container directly to macOS host's Apple Silicon Metal GPU. Fuses Apple's native
MLX machine learning framework with PyTorch, performing zero-copy tensor operations
on Apple's unified memory.

Engine Comparison:
  llama.cpp:   CPU/GPU Hybrid (GGUF) — ~333-345 tokens/sec — short queries, cross-platform
  vllm-metal:  Apple Silicon MLX/Safetensors — Paged Attention + GQA
               Massive context windows, agentic workflows

================================================================================
SECTION 04: PHASE-BY-PHASE CODE ARTIFACTS
================================================================================

PHASE 1 HOST PROVISIONING SCRIPT (bash):
  #!/bin/bash
  apt-get update -y && apt-get install -y qemu-kvm libvirt-daemon-system \
    libvirt-clients bridge-utils ovmf
  systemctl enable --now libvirtd
  sed -i 's/GRUB_CMDLINE_LINUX_DEFAULT="/GRUB_CMDLINE_LINUX_DEFAULT="amd_iommu=on \
    iommu=pt kvm.ignore_msrs=1 /g' /etc/default/grub
  update-grub && update-initramfs -u
  echo "vm.nr_hugepages = 1024" >> /etc/sysctl.conf && sysctl -p

PHASE 2 OPENCORE CONFIG.PLIST (xml):
  <key>PlatformInfo</key>
  <dict>
    <key>Generic</key>
    <dict>
      <key>MLB</key><string>C02XXXXXXXXXX</string>
      <key>SystemSerialNumber</key><string>C02XXXXXXXXXX</string>
      <key>SystemUUID</key><string>XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX</string>
      <key>SystemProductName</key><string>MacPro7,1</string>
      <key>ROM</key><data>AAAAAAAA</data>
    </dict>
    <key>Automatic</key><true/>
  </dict>

PHASE 3 ARANCINI EXECUTION ENGINE (rust):
  fn execute_macos_instruction(aarch64_block: &Block) -> Result<(), ExecutionError> {
      if let Some(native_x86) = ai_cache.get(&aarch64_block.hash) {
          execute_on_bare_metal(native_x86);
          return Ok(());
      }
      let ir = lift_to_llvm_ir(aarch64_block);
      let ir_with_fences = apply_formal_memory_mappings(ir);
      let optimized = llvm_optimize_for_avx512(ir_with_fences);
      let x86_block = compile_to_x86_64(optimized);
      ai_cache.insert(aarch64_block.hash, x86_block.clone());
      execute_on_bare_metal(&x86_block);
      Ok(())
  }

PHASE 4 VIRTUAL GPU KERNEL EXTENSION (cpp):
  IOReturn GlasshouseGPU::SubmitCommandBuffer(
      IOUserClient* client, MetalCommandBuffer* cmd) {
      SerializedMetal payload = SerializeForHost(cmd);
      virtio_ring_write(this->virtio_queue, &payload, sizeof(payload));
      virtio_notify_host(this->virtio_queue);
      return kIOReturnSuccess;
  }

================================================================================
SECTION 05: AI GOD-COMPUTER INTERFACE — MCP
================================================================================

Modern AI models (Claude 3.5/4.5/4.6, GPT-4, Gemini, Manus) do not just chat —
they autonomously operate operating systems. The release of Claude 3.5 Sonnet marked
the definitive transition from conversational LLMs to proactive Large Action Models
(LAMs) via native "Computer Use" capabilities.

THE LIMITATIONS OF VISION-ONLY AUTOMATION:
Initial iterations relied strictly on a "vision-only" methodology — the AI captured
screenshots, analyzed visual layout, and generated X/Y pixel coordinates. This proved
inherently brittle: minor UI updates, resolution scaling changes, notification popups,
or dynamic layouts could derail the agent entirely.

THE MODEL CONTEXT PROTOCOL (MCP) SOLUTION:
MCP is an open-source, standardized transport layer utilizing JSON-RPC 2.0 messages,
developed by Anthropic and standardized by Google and Anthropic in 2025. Supported
by Docker Desktop as of March 2026.

TWO PRIMARY MCP ARCHITECTURES:
  mcp-server-macos-use (mediar-ai, Swift):
    Approach:    Semantic Traversal — traverses AXUIElement accessibility tree
    Capability:  Reads exact UI roles, labels, states — zero hallucination
    Key tool:    macos-use_click_and_traverse with PID targeting
    Coordinates: github.com/mediar-ai/mcp-server-macos-use

  automation-mcp (ashwwwin, TypeScript):
    Approach:    Peripheral Control — raw mouse paths, keyboard chords
    Capability:  Pixel color sampling, window management
    Coordinates: github.com/ashwwwin/automation-mcp

COMPLETE MCP SERVER IMPLEMENTATION (python):
  import asyncio, subprocess, pyautogui
  from mcp.server import Server
  from mcp.types import Tool, TextContent

  app = Server("macos-glasshouse-agent")

  @app.list_tools()
  async def list_tools() -> list[Tool]:
      return [
          Tool(name="click_ui_element",
               description="Moves mouse to X,Y coordinates and clicks.",
               inputSchema={"type": "object",
                            "properties": {"x": {"type": "integer"},
                                           "y": {"type": "integer"},
                                           "click_type": {"type": "string",
                                                          "enum": ["left","right","double"]}},
                            "required": ["x", "y"]}),
          Tool(name="execute_applescript",
               description="Executes native AppleScript to control macOS apps.",
               inputSchema={"type": "object",
                            "properties": {"script": {"type": "string"}},
                            "required": ["script"]}),
          Tool(name="type_text",
               description="Simulates physical keyboard typing.",
               inputSchema={"type": "object",
                            "properties": {"text": {"type": "string"},
                                           "press_enter": {"type": "boolean"}},
                            "required": ["text"]})
      ]

  @app.call_tool()
  async def call_tool(name: str, arguments: dict):
      if name == "click_ui_element":
          x, y = arguments["x"], arguments["y"]
          click_type = arguments.get("click_type", "left")
          pyautogui.moveTo(x, y, duration=0.2)  # Human-like movement latency
          if click_type == "left": pyautogui.click()
          elif click_type == "right": pyautogui.rightClick()
          elif click_type == "double": pyautogui.doubleClick()
          return [TextContent(type="text", text=f"Clicked {click_type} at ({x}, {y})")]
      elif name == "execute_applescript":
          result = subprocess.run(["osascript", "-e", arguments["script"]],
                                  capture_output=True, text=True, check=True)
          return [TextContent(type="text", text=f"Output: {result.stdout}")]
      elif name == "type_text":
          pyautogui.write(arguments["text"], interval=0.01)
          if arguments.get("press_enter", False): pyautogui.press('enter')
          return [TextContent(type="text", text="Text typed successfully.")]

  async def main():
      from mcp.server.stdio import stdio_server
      async with stdio_server() as (read_stream, write_stream):
          await app.run(read_stream, write_stream, app.create_initialization_options())

  if __name__ == "__main__":
      asyncio.run(main())

THE AI LIFECYCLE:
  1. Vision Intake:       AI receives screenshot of macOS desktop via WebRTC or VNC
  2. Reasoning:           LLM analyzes pixels, identifies Xcode icon at (1200, 850)
  3. Action Formulation:  LLM formulates JSON request to MCP Server for click_ui_element
  4. Execution:           MCP server translates JSON into pyautogui commands
  5. Feedback Loop:       New screenshot confirms Xcode opened — loop continues

SECURITY — MICROVM SANDBOXING:
  AWS Firecracker MicroVMs:  Ephemeral, lightweight VMs — spin up in milliseconds
  Kata Containers:           Container-compatible microVM isolation
  Google gVisor:             User-space kernel for lightweight sandboxing
  Zero-trust namespaces:     Strict egress filtering — blocks LAN, whitelists domains
  WebAssembly (Wasm):        Safely sanitizes and executes LLM-generated Python scripts

================================================================================
SECTION 06: DEEP RESEARCH — FACT-CHECKED ANALYSIS (APRIL 2026)
================================================================================

Source: Gemini Deep Research Canvas document — "Deep Research and Fact-Checking"
        31,277 chars — Completed: April 5, 2026, 8:39 AM

THE POST-x86 ERA:
Apple is planning to release macOS 27 exclusively for Apple Silicon, and macOS 28
is slated to remove Rosetta 2 entirely by 2027. The traditional "Hackintosh"
methodology has effectively reached its terminal end of life. Hypervisor-based
virtualization and advanced hardware emulation have become the sole viable pathways.

On native Apple hardware, Apple's Virtualization.framework provides near-bare-metal
execution speeds but is structurally and legally bound to Apple hardware via
specialized entitlements (com.apple.security.virtualization). It cannot be ported
to Linux or Windows host machines.

KOSMICKRISP — VERIFIED:
KosmicKrisp is a real, production-grade component developed by LunarG and packaged
as part of the official Vulkan SDK (version 1.4.335.1 and newer). It is a
sophisticated layered driver that intercepts and translates Vulkan API commands
directly into Metal API commands with minimal translation overhead.

Recent builds of QEMU (v9.2.0 and later) and user-friendly virtualization frontends
like UTM have directly integrated KosmicKrisp alongside the virtio-gpu-gl-pci device.
The introduction of the Apple CoreGL backend has also expanded support for legacy
applications, allowing OpenGL 4.1 acceleration through native translation layers.

Reference: https://www.lunarg.com/the-state-of-vulkan-on-apple-jan-2026/

ARANCINI FRAMEWORK — VERIFIED:
The Arancini framework was presented at the ASPLOS 2026 conference. It is an advanced
Hybrid Binary Translator (HBT) engineered from the ground up to utilize the LLVM
compiler infrastructure, using a proprietary intermediate representation called
ArancinIR.

The most critical scientific contribution is its definitive resolution to the
strong-on-weak memory synchronization problem. Rather than relying on brute-force
memory barriers, the framework utilizes formally verified mathematical mapping
schemes. Benchmark analyses using the Phoenix and EEMBC suites show 81% reduction
in memory access instructions and up to 5x performance improvement over QEMU TCG.

MCP & LAMS — VERIFIED:
The release of Claude 3.5 Sonnet, followed rapidly by versions 4.5 and 4.6 in 2026,
marked the definitive transition of AI from conversational LLMs to proactive Large
Action Models (LAMs) via native "Computer Use" capabilities.

MCP is an open-source, standardized transport layer utilizing JSON-RPC 2.0 messages.
The developer community has produced highly capable open-source MCP servers
specifically for macOS automation. mcp-server-macos-use leverages the native macOS
Accessibility API (AXUIElement) — the AI reads the actual structural UI tree rather
than a flat screenshot.

CONCLUSION FROM DEEP RESEARCH:
The state of macOS virtualization, binary translation, and programmatic system
interaction in 2026 is characterized by the absolute deprecation of legacy x86
paradigms and the rapid maturation of highly sophisticated translation and automation
layers.

The integration of KosmicKrisp within the virtio-gpu paravirtualization stack has
decisively solved the long-standing issue of 3D graphics acceleration across VM
boundaries. Formally verified hybrid translators like Arancini have drastically
mitigated the severe performance penalties associated with translating strong-memory
x86_64 software to run on weak-memory ARM64 and RISC-V architectures.

================================================================================
SECTION 07: THE SCAVENGER ARCHITECTURE — ZERO-COST INFRASTRUCTURE
================================================================================

COMPONENT 01 — THE COMPUTE LOOPHOLE: GitHub Actions (Free Apple Silicon)
GitHub provides free, bare-metal macOS runners (including M1 Apple Silicon) for
CI/CD on public repositories. A workflow requesting runs-on: macos-14 boots a free
M1 instance. Instead of compiling code, the workflow installs VNC, Tailscale, and
enters a sleep 21000 state (5.8 hours). The catch: GitHub kills jobs after 6 hours.
The workaround: a cron job on the Oracle server commits a dummy file every 5.5 hours,
triggering a fresh runner before the old one dies.

COMPONENT 02 — THE PERSISTENT BRAIN: Oracle Cloud Always Free (24GB ARM)
Oracle Cloud offers the most generous free tier in the world: an ARM-based Ampere A1
instance with 4 CPU cores, 24GB RAM, and 200GB block storage for exactly $0.00.
This is the permanent command center. With 24GB RAM, it can run quantized local AI
models (Llama 3, Mistral) via llama.cpp entirely for free.
Execution: Deploy Ubuntu, SSH in, install Docker and Tailscale. Max out the sliders
to 4 OCPUs and 24GB RAM.

COMPONENT 03 — THE NERVOUS SYSTEM: Tailscale Mesh VPN (Free)
Tailscale (free for personal use) links the Oracle server, the GitHub Mac, and the
user's local device into a single virtual LAN. All machines act as if plugged into
the same router, bypassing firewalls and NAT restrictions.
Install via brew install tailscale, authenticate with an ephemeral auth-key stored
in GitHub Secrets.

COMPONENT 04 — THE AI OPERATOR: MCP + Open Source LLMs (Free)
mcp-server-macos-use (Swift, by mediar-ai) intercepts macOS accessibility APIs,
exposing the full UI tree to an AI without requiring expensive vision models.
automation-mcp handles raw mouse/keyboard physics. Combined with a local Llama 3
model on the Oracle server, the entire AI automation stack costs zero dollars per
inference.
Install: git clone https://github.com/ashwwwin/automation-mcp.git
         then bun install && bun run index.ts

GITHUB ACTIONS WORKFLOW (yaml):
  name: Glasshouse Mac Tunnel
  on:
    push:
      branches: [main]
    schedule:
      - cron: '0 */5 * * *'  # Auto-restart every 5.5 hours
  jobs:
    mac-tunnel:
      runs-on: macos-14  # Free M1 Apple Silicon runner
      timeout-minutes: 360
      steps:
        - name: Install Tailscale
          run: brew install tailscale
        - name: Connect to Tailscale network
          run: sudo tailscale up --authkey=${{ secrets.TAILSCALE_KEY }} --ephemeral
        - name: Enable VNC Screen Sharing
          run: |
            sudo /System/Library/CoreServices/RemoteManagement/\
            ARDAgent.app/Contents/Resources/kickstart \
            -activate -configure -access -on \
            -clientopts -setvnclegacy -vnclegacy yes \
            -clientopts -setvncpw -vncpw ${{ secrets.VNC_PASSWORD }} \
            -restart -agent -privs -all
        - name: Install AI Automation Layer
          run: |
            git clone https://github.com/ashwwwin/automation-mcp.git
            cd automation-mcp && bun install && bun run index.ts &
        - name: Hold runner alive (5.8 hours)
          run: sleep 21000

TOS WARNING: The Scavenger Architecture relies on hijacking GitHub Actions for
non-CI/CD workloads. GitHub actively scans for and bans accounts using runners for
remote desktop tunneling. Building startup infrastructure on TOS violations risks
catastrophic, unrecoverable account bans. Use for prototyping only.

================================================================================
SECTION 08: THE BOOTSTRAPPER'S PLAYBOOK — 5 HACKS
================================================================================

Source: "The Glasshouse Directive" whitepaper — Gemini Canvas document, April 2026

HACK 1 — THE ALWAYS FREE MOTHERSHIP: Oracle Cloud
Oracle Cloud offers the most aggressive 'Always Free' tier in the world. Permanently
claim an ARM-based Ampere A1 instance with 4 CPU Cores and 24GB of RAM. Deploy Ubuntu
Linux. Use the Arancini binary translator to run x86 macOS binaries, or host local,
quantized AI models (like Llama 3) via llama.cpp to act as your free, persistent AI
orchestrator.

HACK 2 — EPHEMERAL CI/CD TUNNELING: GitHub Actions
GitHub provides free, bare-metal M1 Apple Silicon runners for public repositories.
Write a GitHub Actions script that boots a macOS runner, installs a mesh VPN
(Tailscale), and initiates a VNC server. You now have a free, hardware-accelerated
Mac. Since jobs time out after 6 hours, write a cron-job on your Oracle mothership
to trigger a new GitHub Action every 5.5 hours, persisting your data via cloud
storage.

HACK 3 — DECOMMISSIONED ENTERPRISE HARVESTING: GSA Auctions
When tech giants and government agencies upgrade, they liquidate massive clusters of
perfectly functional hardware for pennies on the dollar. Monitor platforms like GSA
Auctions (gsaauctions.gov), GovDeals (govdeals.com), and Municibid. You can
frequently purchase pallets of decommissioned Apple Mac Minis or enterprise servers
from universities or federal departments for under $100. Rack them in your garage,
install Proxmox, and build a localized Kubernetes cluster.

HACK 4 — SOVEREIGN AI & CLOUD GRANTS: Billions Available
Governments are pouring billions into 'Sovereign AI' to break US hyperscaler
monopolies. Apply for compute grants:
  EU:            GenAI4EU — massive funding for AI sovereignty
  Canada:        Sovereign Compute Infrastructure Program (SCIP)
  Global South:  EVAH and the AI for Good Impact Awards — up to $60M in compute access
  Microsoft:     Microsoft for Startups — up to $150,000 in Azure credits
  AWS:           AWS Activate — up to $100,000-$350,000 in non-dilutive cloud credits
  EU DMA:        Article 6(7) mandates free interoperability — legal shield for EU startups

HACK 5 — BUG BOUNTY COMPUTE MINING
If you have cybersecurity skills, platforms like HackerOne and Immunefi pay massive
bounties (sometimes up to $1M+). Many programs offer massive multipliers if you accept
payouts in cloud compute credits. Use automated, AI-driven reconnaissance to map
subdomains and mine low-hanging vulnerabilities to indefinitely fund your
infrastructure.

ADDITIONAL APPROACHES:
  DePIN Networks:        Render Network, Fluence, Akash — 45-60% below AWS prices
  Darling:               Wine-like translation layer for macOS Mach-O binaries on Linux
                         github.com/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

STRATEGIC QUOTE: "By weaving these systems together — orchestrating an AI on a free
Oracle server, directing it to execute code on an ephemeral GitHub Mac, while applying
for EU compute grants — you achieve complete digital sovereignty with zero capital."
  — The Glasshouse Directive, April 2026

================================================================================
SECTION 09: LEGAL & LICENSING FRAMEWORK
================================================================================

APPLE EULA POSITION:
Apple's Software License Agreement (SLA) and EULA strictly and unambiguously
stipulate that macOS may only be installed, virtualized, and executed on genuine
Apple-branded hardware. This renders the use of Docker-OSX or QEMU/KVM deployments
running macOS on commodity servers a direct violation of civil contract terms.

APPLE v. CORELLIUM (2019-2023):
A critical legal distinction exists between a EULA violation and actionable copyright
infringement. Corellium successfully commercialized a platform (CORSEC) that
replicated the iOS and macOS kernels for cybersecurity research. Apple sued for
direct copyright infringement.

In a landmark ruling, the 11th Circuit Court of Appeals ruled in favor of Corellium,
establishing that virtualizing an Apple operating system for security research falls
under the "fair use" doctrine. The court found the virtualization software was highly
"transformative" and did not substantially harm Apple's hardware market. The case
culminated in a confidential settlement in late 2023.

LEGAL CATEGORIES:
  Protected (Fair Use):   Cybersecurity research, vulnerability discovery, academic
                          research, security testing in controlled environments
  Prohibited (EULA):      Mass distribution of virtualized macOS for general consumer
                          use, standard software development on commodity hardware,
                          commercial deployment
  EU DMA Shield:          Article 6(7) mandates free and effective interoperability
                          for third parties with iOS/iPadOS hardware/software features
  DMCA 1201(f):           Research exemptions for reverse engineering interoperability

COMPLIANT ALTERNATIVES:
Official cloud computing providers offering macOS hosting (AWS EC2 Mac instances,
MacStadium, Liquid Web) continue to utilize physical, bare-metal Mac Mini hardware
mounted in specialized server racks — ensuring strict EULA compliance.

================================================================================
SECTION 10: ANALYSIS & IMPLICATIONS
================================================================================

THE EMULATION TAX vs. API TRANSLATION:
The single most important conceptual contribution in this dialogue is the "Emulation
Tax" vs. "API Translation" paradigm. Traditionally, running macOS on generic hardware
requires emulating a physical GPU — computationally crushing, resulting in an unusable
interface. Project Glasshouse's insight: stop emulating hardware, start translating
software APIs.

This makes visible that hardware lock-in is increasingly a software enforcement
problem, not a physical physics problem. If a translation layer is fast enough, the
underlying silicon becomes irrelevant. This is the same insight that powered Valve's
Proton (Windows games on Linux) and Apple's own Rosetta 2 (x86 apps on Apple Silicon).

CRITICAL PERSPECTIVES — OPERATIONAL FRAGILITY:
  Legal Risk:          Apple's EULA forbids macOS on non-Apple hardware. Any commercial
                       deployment invites immediate Cease & Desist orders and lawsuits.
  Security:            Bypassing the Secure Enclave and T2 chip compromises
                       cryptographic trust. Storing sensitive data on a 'Ghost Mac'
                       is highly insecure.
  Maintenance:         Apple frequently updates macOS, breaking Hackintosh patches.
                       Maintaining a custom Metal-to-Vulkan bridge requires constant
                       vigilance.
  TOS Violations:      The GitHub Actions tunneling hack violates GitHub's Terms of
                       Service. Building startup infrastructure on TOS violations
                       risks catastrophic account bans.

IMPLICATIONS FOR THE LAUNCHPAD TLP:
For Piqui's domains — education management, entrepreneurship, and social leadership
in Latin America — the implications are profound. The iOS App Store is one of the
most lucrative digital markets in the world, but the entry ticket is a $1,000+ Apple
computer. The Scavenger Architecture proves that the technical barrier to entry can
be bypassed with pure resourcefulness.

The integration of MCP means that AI is no longer just generating text; it is
operating the machine. For a solo founder, an AI agent running on a free cloud server
can act as a tireless employee — scraping competitor data, running tests, managing
social media — all orchestrated through a remote, automated macOS instance.

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.

================================================================================
SECTION 11: COMPLETE RESEARCH COORDINATES & SOURCES
================================================================================

CORE VIRTUALIZATION:
  QEMU/KVM:         https://github.com/qemu/qemu
  Docker-OSX:       https://github.com/sickcodes/Docker-OSX
  OpenCore:         https://github.com/acidanthera/OpenCorePkg
  Proxmox VE:       https://proxmox.com
  gibMacOS:         https://github.com/corpnewt/gibMacOS
  UTM:              https://mac.getutm.app

GRAPHICS & TRANSLATION:
  MoltenVK:         https://github.com/KhronosGroup/MoltenVK
  KosmicKrisp:      https://www.lunarg.com/the-state-of-vulkan-on-apple-jan-2026/
  Darling:          https://github.com/darlinghq/darling
  Arancini HBT:     ASPLOS 2026 conference paper (LLVM-based)

AI & MCP:
  MCP Protocol:     https://modelcontextprotocol.io
  mcp-server-macos: https://github.com/mediar-ai/mcp-server-macos-use
  automation-mcp:   https://github.com/ashwwwin/automation-mcp
  llama.cpp:        https://github.com/ggerganov/llama.cpp
  LangChain:        https://github.com/langchain-ai/langchain
  HyperChat:        Open source MCP client
  Cherry Studio:    Open source MCP client

SECURITY / MICROVM:
  Firecracker:      https://github.com/firecracker-microvm/firecracker
  Kata Containers:  https://github.com/kata-containers/kata-containers
  gVisor:           https://github.com/google/gvisor

FREE CLOUD & INFRASTRUCTURE:
  Oracle Free:      https://cloud.oracle.com/free
  Tailscale:        https://tailscale.com
  GitHub Actions:   https://docs.github.com/en/actions

DECENTRALIZED COMPUTE:
  Akash Network:    https://akash.network
  Render Network:   https://rendernetwork.com
  Fluence:          https://fluence.network

HARDWARE SOURCING:
  GSA Auctions:     https://gsaauctions.gov
  GovDeals:         https://govdeals.com
  Municibid:        https://municibid.com

GRANTS & CREDITS:
  AWS Activate:     https://aws.amazon.com/activate
  MS for Startups:  https://startups.microsoft.com
  EU GenAI4EU:      https://digital-strategy.ec.europa.eu
  AI for Good:      https://aiforgood.itu.int
  EVAH:             https://evah.org

BUG BOUNTIES:
  HackerOne:        https://hackerone.com
  Immunefi:         https://immunefi.com

LEGAL REFERENCES:
  Apple v. Corellium (11th Circuit, 2023)
  EU Digital Markets Act Article 6(7)
  DMCA 17 U.S.C. Section 1201(f)

================================================================================
END OF DOCUMENT
================================================================================

Project Glasshouse Briefing — The Launchpad TLP · thelaunchpadtlp.education
Prepared by Manus AI · April 2026
For research and educational purposes only.
Apple's EULA prohibits running macOS on non-Apple hardware.
All methods described are theoretical and experimental.
The Corellium precedent (11th Circuit, 2023) protects security research use cases only.