Open-Source AI Playground

Build scenes. Train AI.
Run them anywhere.

OpenFluke is a unified playground for physics scenes, native and web runtimes, and AI behaviors. Design once, deploy to Biocraft (web) or Primecraft (native) with identical JSON assets.

Primecraft
Web + Native
Platforms
JSON
Scene Format
LOOM
Open Source AI
Now in Early Access

Primecraft

Explore endless planets with infinite permutations. Train on-device AI companions, build custom levels, and push the limits of your device — all running completely offline.

About
What is OpenFluke?
A complete ecosystem for building, training, and deploying interactive experiences.

Unified Scene Format

Design scenes once using human-readable JSON blueprints. Deploy the same assets to web browsers via Biocraft or native devices via Primecraft without changes.

On-Device Neural AI

Train real neural networks directly on your device with our LOOM framework. No cloud required — your AI companions learn and evolve locally.

Physics-First Design

Every interaction is grounded in real physics simulation. Constraints, collisions, and forces behave consistently across all platforms.

Open Source AI Framework

LOOM (our neural AI engine) and IsoCard (scene format) are fully open source. Build, extend, and integrate them into your own projects.

Gallery
Primecraft Gallery
Snapshots from the native runtime — high frame rates, zero drift.
About Primecraft
Status
Platform Parity Tracker
Feature synchronization between Biocraft (web) and Primecraft (native).
Development Progress
Tracking feature parity across runtimes
Biocraft (Web)
IsoCard
Scene Gen
Constraints
Physics Glue
Reflex: Auto
Automation
Reflex: Seq
Timeline
Possession
Input Control
AI Brain
Loom / Paragon
MIGRATING
Primecraft (Device)
IsoCard
Scene Gen
Constraints
Physics Glue
INTEGRATING
Reflex: Auto
Automation
Reflex: Seq
Timeline
Possession
Input Control
AI Brain
Loom / Paragon
Workflow
How It Works
One pipeline from authoring to AI training to deployment.
1

Design in Your Lab

Sketch scenes, physics, and controllers with JSON blueprints. Keep everything versioned and portable.

2

Train LOOM AI

Train operators locally or in the browser with LOOM; export reflexes and networks that sync with your scenes.

3

Deploy Anywhere

WebGPU studio for instant iteration; native runtime for shipping, tournaments, and zero-drift QA.

4

Share & Publish

Publish scenes publicly; anyone can pull JSON over the API or launch the interactive view.

Open Source Core

LOOM (our neural AI framework) and IsoCard (our scene format) are fully open source. Explore the code, contribute features, or integrate them into your own projects.

Discover
Random Scenes from the Lab
A fresh shuffle of published scenes every load. Click through to explore.
Shuffle
sandbox

course_1

By @sam

55 views • 2 weeks ago

sandbox

reflex-sequence-lets-go-bowl

By @sam

129 views • 1 month ago

bowl
sandbox

chain_reaction

By @sam

43 views • 1 week ago

crash
sandbox

bear-or-dog

By @sam

123 views • 1 month ago

sandbox

wall_n_ball

By @sam

36 views • 1 week ago

crash
sandbox

bear_or_dog

By @sam

103 views • 1 month ago

beardog
operator

Treasure Hunt Explorer: Q-Learning Maze

By @ailearner

121 views • 1 month ago

BeginnerNavigationQ-LearningTutorial
sandbox

Ground_Arena

By @sam

30 views • 1 week ago

crash
sandbox

playground

By @sam

87 views • 3 weeks ago

playground
sandbox

reflex-automation-example

By @sam

212 views • 1 month ago

reflex
sandbox

human_with_ai

By @sam

232 views • 1 month ago

aitrained
sandbox

jenga_tower

By @sam

39 views • 1 week ago

crash