Godot 4-powered simulation with 27,000 procedural planets providing diverse testing environments for AI agents through TCP connections.
Pioneering the future of artificial intelligence through distributed computing, immersive simulation, and evolutionary neural networks. Where Cutting-Edge Meets Reality.
Three interconnected services powering the next generation of AI research
graph TD
OF[OpenFluke AI Platform]
BF[BioFoundry Simulation]
PG[Paragon Framework]
TR[TREE Evolution Engine]
PL[27k Procedural Planets]
TCP[TCP Connections]
AG[AI Agent Control]
WG[WebGPU Acceleration]
TA[Type Agnostic Support]
BR[Cross Platform]
GF[Growth Function]
MN[Micro Networks]
CP[Checkpoint Sampling]
DC[Decentralized Computing]
ED[External Devices]
EX[Experimenters]
OF --> BF
OF --> PG
OF --> TR
BF --> PL
BF --> TCP
BF --> AG
PG --> WG
PG --> TA
PG --> BR
TR --> GF
TR --> MN
TR --> CP
TR --> DC
TR --> PG
AG --> TR
DC --> ED
EX --> BF
MN --> CP
GF --> MN
style OF fill:#ff6b6b,stroke:#333,stroke-width:3px,color:#fff
style BF fill:#4ecdc4,stroke:#333,stroke-width:2px,color:#fff
style PG fill:#4ecdc4,stroke:#333,stroke-width:2px,color:#fff
style TR fill:#4ecdc4,stroke:#333,stroke-width:2px,color:#fff
BioFoundry
Simulation Environment
Godot 4-powered simulation with 27,000 procedural planets providing diverse testing environments for AI agents through TCP connections.
Paragon
AI Framework
Type-agnostic Go framework with WebGPU acceleration, supporting multiple data types and cross-platform deployment.
TREE
Evolution Engine
Distributed training system that evolves neural networks through micro-network extraction and decentralized computing.
Revolutionary distributed AI training that transforms idle devices into a planetary-scale neural network. Watch as micro-networks evolve, compete, and improve autonomously across phones, laptops, and edge devices.
Cost Reduction
vs Traditional Cloud Training
Faster Iteration
Distributed Micro-Training
Scalability
Every Device is a Worker
Open Source
Apache 2.0 Licensed
Paragon extracts micro-networks from checkpoint layers, creating trainable sub-models that preserve the original network's behavior.
Micro-networks are sent to available devices (phones, laptops, edge devices) for parallel improvement attempts using different architectures.
Each device experiments with adding layers, changing activations, and optimizing weights. The best improvements are reintegrated into the main network.
Micro-Network Surgery
Surgically extract and modify specific neural network segments without losing model integrity or performance.
ADHD Scoring
Advanced performance evaluation using Accuracy Deviation Heatmap Distribution for precise model assessment.
Distributed Training
Leverage idle computing power across phones, laptops, and edge devices for massive parallel processing.
Privacy-First
Data never leaves your infrastructure. Training happens on distributed checkpoints, not raw data.
Real-time Evolution
Networks grow and adapt in real-time as they encounter new data, continuously improving performance.
Type-Generic Framework
Works with float32, float64, and even integer neural networks. WebGPU acceleration included.
Watch as neural networks evolve in real-time across distributed devices
Current Demo: MNIST Classification
Network Layers: 5
Active Devices: 12
Accuracy: 97.3%
Growth Events: 8
Training Cost: $0.12
vs Cloud Cost: $12.50