OpenFluke Documentation

Learn how to build physics scenes, train AI behaviors, and deploy across web and native platforms using the OpenFluke ecosystem.

What is OpenFluke?

OpenFluke is an ecosystem for building physics-based simulations where humans and AI can learn, play, and create together. The platform consists of interconnected tools that share the same JSON scene format and AI model interfaces.

One Ecosystem, Multiple Environments

Design in your Lab, test in Biocraft (web studio) or Primecraft (native app), train with Paragon AI Framework. All using the same JSON scenes and models.

Core Components

Your Lab (Creator Space)

Your personal workspace for designing scenes, training AI operators, and experimenting with physics simulations. The Lab is where creation begins.

Biocraft (Web Studio)

A browser-based, open-source sandbox for designing levels, experimenting with AI, and training behavior models. Built with TypeScript + Ionic, powered by IsoCard for physics and scene management.

Primecraft (Native Platform)

A native Godot 4.5 (C#) platform that runs and distributes worlds at scale, with multiplayer, benchmarking, and AI reproducibility tools. Loads the same JSON scenes as Biocraft.

Paragon (AI Framework)

A portable AI framework with WebGPU acceleration and deterministic training. The same model runs identically in browser (WASM) and native (C-ABI) environments. Powers all AI operators in the ecosystem.

How It Works

The OpenFluke ecosystem follows a simple but powerful workflow that keeps everything connected:

Flow

Design → Train → Deploy → Benchmark

1. Design scenes in Lab or Biocraft (JSON format)
2. Train AI operators with Paragon framework
3. Deploy to Biocraft (web) or Primecraft (native)
4. Benchmark reproducibility and performance

Key Features

Shared JSON Scene System: One scene definition works everywhere. Deterministic physics using Jolt (via IsoCard).
Controller Layer: Players and AI share the same control interface. Live swap between human and AI during simulation.
AI Operators: Reinforcement learning templates that run identically in browser and native. Reproducible results across devices.
Scheduler System: Time-based and event-based automation for dynamic scenes and gameplay logic.
Cross-Runtime AI: Train in browser, run on mobile, benchmark on desktop. Paragon ensures bit-for-bit reproducibility.

Proof of Ecosystem

This MVP demonstrates that AI and human control can coexist fluidly, AI models can deploy and reproduce results across devices, and creators can design without engine lock-in.

Installation

JavaScript/TypeScript

bash

npm install @openfluke/portal

Python

bash

pip install paragon-py

Go

bash

go get github.com/openfluke/paragon/v3

C/C++ (via Teleport)

bash

git clone https://github.com/openfluke/teleport
cd teleport
go build -buildmode=c-shared -o libteleport.so

Platform Support

WebGPU acceleration requires a compatible GPU and browser/runtime. CPU fallback is automatic. See the system requirements for details.

Next Steps

Now that you understand the basics, here's where to go next:

Quick Start Tutorial

Build your first scene in 5 minutes

Core Concepts

Deep dive into architecture

Train Your First AI

Create an autonomous operator

Example Scenes

Explore community creations

Need Help?

If you can't find what you're looking for in the documentation, here are some other resources:

Join our Discord community for real-time help
Check out GitHub Discussions for Q&A
Browse example projects on GitHub
Report bugs on the issue tracker

Contributing

OpenFluke is open source! We welcome contributions of all kinds. Check out our contributing guide to get started.