Maya-OS: An Affective Spiking Neural Network as a Conversational Operating System Arbitration Layer

We present Maya-OS, the first implementation of a four-neuron affective Spiking Neural Network deployed as a real-time conversational operating system arbitration layer. Four LIF neurons — Bhaya (fear, τ=3), Vairagya (wisdom, τ=20), Shraddha (trust, τ=10), and Spanda (aliveness, τ=5) — continuously read live system metrics and conversational intent signals, producing an affective voltage vector that governs process scheduling, resource protection, and syscall arbitration decisions without hardcoded rules. Across five experimental sessions (133 ticks, Windows 11), we demonstrate autonomous threat accumulation, Shraddha-mediated escalation suppression as an emergent safety primitive, confirmed LIF action potential discharge, intent-driven neural modulation, and autonomous homeostatic recovery. No prior work has deployed an affective SNN as a live OS arbitration layer. Extends Swaminathan (2026) https://doi.org/10.5281/zenodo.19151563. Codebase: https://github.com/venky2099/Maya-OS