WhiteTowerAI/ohmygame

What it solves

Making a game typically requires juggling multiple disconnected tools for design documentation, 2D art, 3D modeling, video production, and coding. OhMyGame consolidates these into a single application where an AI agent handles the heavy lifting of asset creation, implementation, and testing.

How it works

The system is built around a central canvas and an integrated AI agent powered by Pi.

• The Canvas: A visual workspace where design docs, concept art, 3D models, and videos coexist. Assets can reference each other, allowing a sketch to evolve into a rigged 3D model.
• The Agent: An AI that drives the canvas by drafting documents, generating assets, writing code, and performing automated playtesting. It treats game actions (like playing the game or generating media) as tool calls.
• File-Based Workspace: Everything is stored as standard files (Markdown for docs, JSON for boards, standard web projects for games), ensuring the agent can edit them like code.
• BYOK (Bring Your Own Key): Users connect their own API keys from 40+ providers for LLMs, image, video, and 3D generation.

Who it’s for

Solo developers or "one-person game studios" who want to rapidly prototype and build games, interactive stories, or web-based applications using AI.

Highlights

• End-to-end automation: The agent can design, build, playtest, and fix bugs in a loop.
• Multimodal asset pipeline: Integrated support for generating images, video, and rigged 3D models.
• Engine flexibility: Supports web games (Three.js, React Three Fiber, Phaser) and interactive stories via a node-graph architecture.
• One-click publishing: Builds games into a shareable link playable in any browser.
• Provider agnostic: Works with a wide range of models including Anthropic, OpenAI, DeepSeek, and Qwen.