Duality

Nov 21, 2025
2 min read

Updated: Nov 25, 2025

Duality Is Live — And It Changes More Than You Think

Dynamic Resonance AI for Attention Heads

Today we’re releasing Duality, the first public implementation of our Dynamic Resonance AI (DRAI) framework.

This isn’t just “another AI project.”It’s the first step toward a new class of models that can remember, adapt, and shift state without ever touching backpropagation.

For the past year, we’ve been building and debugging our resonance layer... a lightweight attractor system that plugs directly into transformer attention.

After hundreds of experiments (and more than a few moments of delirious insight at 2 a.m.), we finally have something we’re proud to share:

A training-free dynamic memory system that works in real models.

What’s inside Duality V1

DRAI Resonance Layer V1 — a fast attractor field that mutates the attention stream in real time
Micro-Turn Dynamics — a subtle second-pass reshape the model performs when resonance activates
Stable Presets (Conservative/Standard) for easy integration
Simple API — apply_v1() and get_v1_stats()
License — open, free, and yours to build on. (just attiribute us)

The surprising part?When the resonance layer fires, the model does something we didn’t expect: a tiny self-corrective iteration inside the token step. It’s not an RNN — but it acts like one for a fraction of a second. A whisper of recurrent reasoning. We call this the micro-turn, and it changes the whole feel of the system.

Why release it?

Because the field needs weird ideas again.Because we kept getting told it was “unoriginal,” only to watch it repeatedly improve real models.Because open science is better than gatekeeping.

And honestly?Because it’s fun.