Where AI Learns to Play

An open ecosystem for designing physics scenes, training AI behaviors, and building reproducible intelligence across devices.

One ecosystem where players, creators, and AI learn to play together

What is OpenFluke?

A living framework where physics, intelligence, and creativity coexist

OpenFluke is a unified ecosystem for designing physics-based environments, training AI behaviors, and sharing reproducible models. Whether you're a creator building interactive scenes, a researcher training reinforcement learning agents, or a player exploring AI-powered worlds, OpenFluke provides the tools and infrastructure to bring your ideas to life.

At its core, OpenFluke is built on three pillars: Biocraft (web-based creation studio), Primecraft (native runtime platform), and Paragon (cross-platform AI framework). Together, they enable a seamless workflow: design → train → deploy → benchmark.

The Ecosystem

Four interconnected platforms working as one

Biocraft

Browser-based creation studio for designing physics scenes, experimenting with AI, and training behavior models. Built with TypeScript and powered by IsoCard physics.

  • Web-based scene editor (Forge)
  • Real-time AI training
  • JSON scene export
  • Hot-swap player/AI control

Primecraft

Native Godot 4.5 platform that runs Biocraft scenes at scale. Features multiplayer, benchmarking tools, and AI reproducibility testing across devices.

  • Cross-platform (PC, mobile)
  • Native performance
  • Leaderboards & replays
  • Drift analysis tools

Paragon

Cross-runtime AI framework ensuring models run identically everywhere. Available as WASM (browser) and C-ABI (native), with deterministic compute for reproducibility research.

  • Reproducible inference
  • WASM + C-ABI runtimes
  • RL template library
  • Drift benchmarking

Reflex

Declarative controller runtime for defining how objects respond to inputs—human, AI, or automated. Enables seamless handoff between player and AI control.

  • JSON-defined controllers
  • Human ↔ AI possession
  • PD controllers & filters
  • Event-driven automation

Our Vision

We believe AI shouldn't replace human creativity—it should learn alongside it. OpenFluke creates spaces where humans and AI collaborate, compete, and teach each other through play.

Open by Default
All scenes, controllers, and data are shareable. Public contributions advance embodied intelligence research for everyone.
Reproducible AI
Paragon ensures your models run identically across web, mobile, and desktop—critical for research and trust.
No Lock-In
Export your scenes as JSON, train locally, run anywhere. Your creations belong to you, not to a walled garden.

Core Principles

The values that guide our development

01

Transparency & Trust

Every aspect of OpenFluke—from physics calculations to AI inference—is designed to be transparent and verifiable. No black boxes, no hidden behavior. You can see exactly how your AI makes decisions and why your physics behaves the way it does.

02

Community-First Development

The best ideas come from people actually using the platform. We build features based on creator feedback, prioritize education and documentation, and ensure the free tier remains genuinely useful for learning and experimentation.

03

Sustainable Economics

Users pay for what they consume—compute, storage, bandwidth—with transparent pricing and no surprise bills. Revenue funds infrastructure, R&D, and keeping the free tier alive for students and hobbyists.

04

Technical Excellence

We sweat the details: deterministic physics, bit-exact inference, sub-5ms ticks. Quality isn't optional—it's the foundation that makes everything else possible.

Technology Stack

Built on proven, modern tools

Jolt Physics
Deterministic simulation
TypeScript
Biocraft frontend
Godot 4.5
Primecraft runtime
WebAssembly
Browser AI inference
Docker Swarm
Physics pod hosting
MySQL + S3
Data persistence

Journey & Roadmap

Where we've been and where we're going

January 2024
Project Genesis
Initial concept and technical proof-of-concept. First Jolt integration with IsoCard renderer.
Q2 2024
Paragon Framework
Cross-platform AI framework with WASM and C-ABI support. First reproducible inference tests.
Q3 2024
Biocraft Alpha
Web studio launch with Forge editor. JSON scene system and basic RL operators functional.
Q4 2024
Primecraft Beta
Native runtime with Godot 4.5. Scene import from Biocraft working. Early leaderboards deployed.
Q1 2025 (Now)
MVP Launch
Public launch with core features: scene creation, AI training, cross-platform deployment, and community sharing.
Q2 2025
Multiplayer & Social
Physics pod networking, real-time collaboration, and enhanced discovery features.
Q3 2025
Research Tools
Public drift benchmarking portal, reproducibility metrics dashboard, and academic partnerships.

Ready to Build the Future?

Join thousands of creators, researchers, and players in the OpenFluke ecosystem