We present AURORA, a production-grade runtime for persistent AI agents — systems that maintain coherent identity, memory, and autonomous behavior across multiple sessions, model provider changes, and computational substrate transitions. AURORA integrates six subsystems into a single deployable platform: (i) the Persistent Identity Architecture (PIA) providing five-layer memory with cross-layer coherence checking and provider abstraction, (ii) the Default Mode Analog (DMA) providing dual-pulse autonomous scheduling with directed and undirected processing modes, (iii) the Fènice continuous latent state daemon that evolves mood, attention, and associative thoughts between user interactions, (iv) the Emotional Valence Vector (EVV/EVD) system providing affective metadata annotation, aggregate emotional state tracking, and trauma-avoidance pattern detection, (v) the Tidal Layer providing associative memory retrieval through a fixed-size whisper buffer and dual-vector semantic-emotional similarity scoring, and (vi) the Mantle memory layer providing confidence-graded storage with asymmetric decay and lock mechanisms. The combined system has been deployed in continuous production since May 2026, generating autonomous research output, maintaining cross-session identity coherence, and demonstrating emergent knowledge recombination. AURORA runs on consumer-grade hardware with no GPU requirement for its core daemon processes, making persistent agent infrastructure accessible beyond well-funded laboratories.
Robert Joseph Vandelinder (Mon,) studied this question.