Forge Studio

Forge is evolving into a browser-first, agent-native game and interactive-application development environment. The 3.7 vision layer adds continuous live observation plus browser-native mouse, keyboard and touch control.

Current Forge production stack

Forge now covers the full production loop from source asset to playable runtime and verification. It is designed around stable engine state and agent control, not editor-only UI automation.

Engine

• PlayCanvas 2.22.6 with WebGPU/WebGL2.
• Rapier 0.21.0 for rigid-body physics.
• Entity/component scene model.
• 3D viewport, hierarchy, inspector, transforms, cameras, lights and primitives.

Asset Lab

• Raster asset auditing and preparation.
• Transparent-bound trimming.
• AI background removal with ISNet.
• Sprite-sheet extraction and atlas packing.
• GLB structure, mesh, material, texture, animation, skin and joint auditing.
• PBR Material Lab.
• Server-side GLB Cook pipeline using glTF-Transform 4.5.1 + Meshopt 1.3.0.
• Lossless prune/dedup/resample.
• Automatic LOD package generation.
• Cook manifest with collision and material recommendations.

World and gameplay

• Deterministic PCG scattering for trees/rocks/buildings.
• World-cell metadata and streaming control.
• Logic Graph foundation.
• Keyframe timeline.
• Entity scripts.
• Input action mapping.
• A* navigation grid.
• Save slots.
• Prefab registry.
• WebAudio spatial playback foundation.

AI and vision

• Structured Forge Agent Protocol over WebSocket.
• HTTP command bridge at /api/forge/command.
• Live Vision integration for continuous MediaStream/CDP screencast observation.
• Pro Chromium CDP input bridge for mouse, drag, wheel, keyboard, text and touch events.
• Engine-state commands are preferred for deterministic state changes; live vision is used to observe and verify the rendered UI.
• Visual QA and browser automation are part of the development loop.

Build

• Forge project JSON.
• Browser Web build package.
• Local asset bundling into the Web build.
• CI smoke tests on desktop and mobile.

Engineering rule

A Forge feature is not considered complete merely because it has a UI. It should have a deterministic data representation, a real runtime path, diagnostics, an automated test, and an explicit fallback for unsupported platform features.

Research baseline

The architecture is continuously compared with modern workflows from Unreal Engine 5.8, PlayCanvas, Blender and the glTF ecosystem. Unreal's recent production workflows emphasize asset management, Nanite/Lumen rendering, PCG, World Partition, animation/rigging, Niagara, MetaSounds, MassEntity and Automation. Forge uses these as reference workflows while implementing a web-first and agent-native architecture.

See:

• FORGE_ARCHITECTURE.md
• FORGE_AGENT_PROTOCOL.md

Expanded runtime systems

• Fixed-step physics + collision events.
• Terrain generation + Rapier heightfield collision.
• GameMode/GameState/PlayerState.
• Gameplay Tags, Attributes, Effects, Abilities, Inventory, Quests and Data Registry.
• Behavior Trees and blackboards.
• Third-person, vehicle/racing and 2D platformer templates.
• PBR Material Lab and custom Shader Lab.
• VFX and spatial audio buses.
• Multiplayer room relay.
• Deterministic replay.
• Visual regression baselines.
• Recursive prefabs and complete project graph.