An in-house Desktop 3D Game Development Studio and AI Harness tailored for Android.
Bridging the visual agility of GDevelop and the component depth of Unity, powered by an integrated AI Copilot and autonomous mobile playtesting via Google Artemis.

🌟 Key Highlights

3D Engine Core: Three.js PBR graphics and @dimforge/rapier3d-compat WebAssembly physics with rigid bodies, colliders, raycasting, and mobile character controllers.
Hybrid Scripting Paradigm:
- Visual Event Sheets (GDevelop style): Conditions & Actions editor for no-code logic and rapid prototyping.
- TypeScript Components (Unity MonoBehaviour style): Full lifecycle (awake, start, update(dt), onCollisionEnter, onDestroy).
Desktop Studio Interface:
- 3D Viewport: Orbit controls, camera presets, transform gizmos (Translate, Rotate, Scale), and real-time mobile touch emulation.
- Scene Hierarchy: Tree view of GameObjects, parent-child nesting, and primitive spawners.
- Inspector: Live editing of transforms, PBR materials, lights, physics bodies, and mobile joysticks.
- Asset Browser: Project management for glTF 2.0 (.glb/.gltf), textures, audio, and scene files.
AI Studio Copilot:
- Natural language prompt-to-world ("Create a desert obstacle course with 5 pillars, floating platforms, sunset lighting, and a player sphere").
- Automated code generation, component wiring, and self-healing.
Google Artemis Autonomous Game QA:
- Integrates with Google Artemis (99%+ AndroidWorld SOTA).
- Headless engine runs: boots scenario specs on the real runtime (Rapier3D + EventSheet) — real sim FPS, frame time, GPU draw-call estimate against the 100-call mobile budget, memory heap, and game-rule assertions.
- Regression gate: every run compares against the same-scenario baseline in project_memory.sqlite and emits SUCCEEDED / REGRESSED / FAILED.
- On-device playtests: autonomous touch automation on real Android phones or emulators via ADB; audits frame time and diagnoses Logcat crashes.
Two-Tier Model Context Protocol (MCP):
- Built-in MCP server (harness/mcp_server.py) allowing external AI coding assistants (Antigravity, Claude Code, Cursor) to inspect the 3D scene and execute editor actions programmatically (9 tools).
- Scene mutations persist to harness/scenes/active_scene.json and sync snapshots into project memory; studio_run_artemis_qa boots that exact live scene headless.

🚀 Quick Start

1. Prerequisites

Node.js (v18+ or v20+)
Python (3.10+)
Android SDK & ADB (configured in $ANDROID_HOME with USB debugging enabled on phone/emulator)

2. Installation

# Clone repository
git clone https://github.com/Heretek-Games/Heretek-Engine.git
cd Android-Dev-Studio

# Install monorepo dependencies
npm install

# Build core 3D engine
npm --workspace=engine run build

3. Launch Desktop Studio

# Start Vite development server
npm run dev

Open http://localhost:3000 in your browser (or run with Electron).

🧪 Testing

# Run Engine ECS, Scene Graph, and EventSheet Unit Tests
npm test

🤖 Google Artemis Autonomous QA

Headless scenario runs (default, deterministic) — boots the scene on the real engine runtime with Rapier3D physics and EventSheet execution, then evaluates game rules and compares metrics against the same-scenario baseline:

# QA the live MCP-controlled scene (harness/scenes/active_scene.json)
python3 harness/agents/artemis_qa_runner.py --goal "Verify player physics and spin events"

# QA a named scenario spec
python3 harness/agents/artemis_qa_runner.py --goal "Mini arena QA" --scenario harness/config/scenarios/mini_arena.json

Verdicts: SUCCEEDED · REGRESSED (FPS −20% / frame time +20% / draw calls +25% / heap +30% vs same-scenario baseline) · FAILED (rule assertions). Reports land in harness/artemis_report.json. The Studio's Artemis dock drives the same runner through the dev-server bridge (POST /api/qa/run).

On-device playtests on a connected Android phone or emulator via ADB (touch automation and frame-time/Logcat auditing) follow the locator pattern in harness/config/artemis_game_rules.md.

To connect AI coding assistants to the Studio via Model Context Protocol:

python3 harness/mcp_server.py

📁 Repository Structure

Android-Dev-Studio/
├── app/                              # Desktop Studio UI (React, Tailwind, Three.js Viewport)
│   ├── src/components/
│   │   ├── Viewport3D.tsx            # 3D canvas, gizmos, mobile touch HUD
│   │   ├── Hierarchy.tsx             # Scene graph tree
│   │   ├── Inspector.tsx             # Transform, materials, physics inspector
│   │   ├── EventSheetEditor.tsx      # GDevelop visual condition-action editor
│   │   ├── AIHarnessDock.tsx         # Natural language world & logic copilot
│   │   ├── ArtemisQADock.tsx         # Google Artemis autonomous test runner
│   │   ├── AssetBrowser.tsx          # glTF models, textures, audio
│   │   └── DeviceBar.tsx             # ADB device selector & APK deployer
│   └── vite.config.ts
│
├── engine/                           # Core 3D Game Engine (Shared Editor & Android Runtime)
│   ├── src/core/                     # Transform, Component, GameObject, Scene, EngineContext
│   ├── src/components/               # MeshRenderer, LightComponent, RigidBody3D, Collider3D, MobileController
│   ├── src/events/                   # EventSheet visual condition-action interpreter
│   ├── src/physics/                  # Rapier3D WASM physics world
│   └── src/input/                    # Mobile touch joystick & keyboard mappings
│
├── templates/
│   └── android-container/            # High-performance Android Gradle Template
│       ├── app/src/main/
│       │   ├── AndroidManifest.xml   # Fullscreen landscape, WebGL2 hardware acceleration
│       │   └── java/                 # MainActivity with WebViewAssetLoader & haptic bridge
│       └── build.gradle
│
└── harness/                          # AI Harness, Artemis QA & Studio MCP Server
    ├── agents/artemis_qa_runner.py   # Autonomous mobile playtest runner
    ├── mcp_server.py                 # Studio Model Context Protocol server
    └── config/artemis_game_rules.md  # Autonomous QA behavioral rules

README.md Unescape Escape

Heretek 3D Android Studio & AI Harness