Godogen is a code generation tool that autonomously builds games for Godot, Bevy, and Babylon.js using Claude Code or Codex as the agent.
The tool solves the problem of game development requiring extensive manual coding and asset creation by automating the entire pipeline. You describe a game in natural language, and the agent generates a complete game repository with project scaffolding, engine-specific code, and assets. The agent works by reading the running game's actual output—either a live URL or recorded video—and iterating based on visible defects rather than assuming correctness from compilation. Asset generation uses multiple specialized models: Gemini for character references, xAI Grok for textures and simple objects, and Tripo3D for 3D models and rigged animation. For Godot 4 it generates C# projects with scene generation and Jolt physics; for Bevy it creates Rust ECS scenes with offscreen capture; for Babylon.js it produces TypeScript games served live. You can watch the live game and steer at decision points, or run unattended and receive a proof recording.
This approach suits developers exploring rapid game prototyping or those wanting to delegate implementation details to an agent while retaining creative direction. The tool requires significant infrastructure: Godot 4 with .NET, Rust and Cargo, Node.js 22.12 or later, Chrome with WebGL2, Python 3, and API keys for Google AI, xAI, and Tripo3D. It also depends on system packages including Vulkan tools, xvfb, ffmpeg, and ImageMagick, and has been tested on Ubuntu, Debian, and macOS. The agent's output is a thin published repository containing only a runtime manifest, a one-page engine guide, and the asset-generation skill; everything else is recreated from the guide at runtime.
Development activity shows consistent iteration on the core generation pipeline with focus on proof-based validation rather than static analysis. The project maintains support across three distinct game engines with engine-specific guides and rendering logic, indicating sustained effort to keep multiple code paths functional. Integration with external AI services for asset generation suggests ongoing refinement of which models work best for different asset types. The tool is designed for server deployment with multi-hour generation runs, reflecting practical experience with long-running autonomous processes.
