I built an open-source self-learning neuromorphic engine in C++17. This short video walks through what it does, how it learns, and where it currently stands.
What it is
FractalBrainOS is a hierarchical oscillatory neural network. Its substrate is a fractal topology of triplets — groups of three phase oscillators that synchronize via the Kuramoto model and recursively form higher-level triplets to arbitrary depth.
Learning combines three mechanisms:
- STDP with free-energy gating
- A reduced-form implementation of Friston's free energy principle
- A seed implementation of active inference
What works today
- Hierarchical Kuramoto synchronization across all levels
- STDP with free-energy gating
- Delay-line memory (standing-wave storage)
- Sleep consolidation with structural pruning
- Homeostasis to prevent pathological synchronization
- int16 phase quantization, LUT-accelerated sin/cos
- UDP daemon mode for external stimulation
What awaits integration
- VSA (Vector Symbolic Architecture)
- Full active inference policy
- P2P networking
- LLM-in-binary integration
The README documents exactly what works and what is a hook. No overclaiming.
The engine metaphor
It is a digital engine, not a product. Like a combustion engine — it works, but what you build around it is up to you. Connect sensors and motors, and it can learn to control a drone. Connect an LLM, and it can speak.
The code
GitHub: https://github.com/NineNi999neNine/FractalBrainOS
MIT license. Feedback welcome — even critical.
Top comments (0)