CortexAI

White-box, CPU-first neurodynamical computing for AI applications.

CortexAI is an open-source alternative to black-box LLMs, built on established neurodynamical models (Kuramoto, Wilson-Cowan, Izhikevich, Lyapunov, Schultz/dopamine) rather than massive statistical learning.

Key Features

• Total Explainability: Every output can be traced back to the equations that produced it.
• Lightweight: CPU-only, low memory footprint, low latency.
• Modular: Each equation is an independent module, composable in a graph.
• Scientific Rigor: All performance metrics (latency, RAM) are reproducible and publicly benchmarked.

Quick Start

Python

pip install cortexai

from cortexai import Kuramoto, CognitiveGraph

# Create a Kuramoto oscillator network
kuramoto = Kuramoto(num_oscillators=10, coupling_strength=1.0)

# Run simulation
for step in range(100):
    state = kuramoto.step(dt=0.01)
    print(f"Step {step}: order parameter = {kuramoto.order_parameter():.4f}")

Rust

use cortexai_core::kuramoto::{Kuramoto, KuramotoParams};

let params = KuramotoParams {
    k: 1.0,
    num_oscillators: 10,
    natural_frequencies: vec![1.0; 10],
};

let mut kuramoto = Kuramoto::new(params);
let state = kuramoto.step(&input, &params, 0.01);

Architecture

┌─────────────────────────────────────────────┐
│  Frontend (Next.js / React)                  │
│  Dashboard temps réel + Playground            │
└───────────────────▲───────────────────────────┘
                    │ WebSocket / REST / SSE
┌───────────────────┴───────────────────────────┐
│  API / Backend (FastAPI)                      │
│  Auth · Rate limiting · Observabilité          │
└───────────────────▲───────────────────────────┘
                    │ bindings (PyO3)
┌───────────────────┴───────────────────────────┐
│  Cognitive Layer (Python)                     │
│  Orchestrateur du graphe d'équations           │
│  Config déclarative (YAML/JSON)                │
└───────────────────▲───────────────────────────┘
                    │
┌───────────────────┴───────────────────────────┐
│  Core Engine (Rust)                           │
│  Kuramoto · Wilson-Cowan · Izhikevich          │
│  Lyapunov · Dopamine · Solveur ODE             │
└─────────────────────────────────────────────────┘

Modules

Module	Description	Reference
Kuramoto	Coupled oscillator synchronization	Kuramoto, 1975
Wilson-Cowan	Excitatory/Inhibitory population dynamics	Wilson & Cowan, 1972
Izhikevich	Spiking neuron model	Izhikevich, 2003
Lyapunov	Stability analysis	Lyapunov, 1892
Dopamine	Reward Prediction Error	Schultz, 1997

Performance

Metric	Target	Status
Latency (100 oscillators)	< 88ms	✅
Memory (1000 neurons)	< 231 MB	✅
CPU-only	Yes	✅

Contributing

See CONTRIBUTING.md for guidelines.

License

MIT OR Apache-2.0

Citation

@software{cortexai2026,
  title = {CortexAI: White-box Neurodynamical Computing},
  author = {CortexLLMs},
  year = {2026},
  url = {https://github.com/CortexLLMs/cortexai}
}