Effective dementia care is hindered by fragmented perspectives among patients, caregivers, and clinicians, leading to delayed or suboptimal interventions. We introduce a novel Ambient Assisted Living (AAL) framework that reconceptualizes the home as an active caregiving partner, aiming at safe aging in place. Our key contribution is a privacy-preserving, affordable, holistic system with continuous real-world validation, designed for unobtrusive, zero-interface operation. The system integrates novel low-cost hardware solutions, including a remote audio announcement device, a bidirectional intercom, and a “Zero-Interface Mirror” enabling non-recorded, real-time video co-presence between patients and guardians. A local edge-based AI architecture processes multi-modal inputs without persistent data storage, ensuring privacy by design. The system combines real-time acoustic event detection (via Audio Spectrogram Transformer), automatic speech recognition and diarization, dual-stage fall detection, and extraction of acoustic biomarkers from natural conversations, which are selectively uploaded -depending on Dementia stage- to a clinician-facing dashboard for longitudinal monitoring. A 10-month in-home deployment demonstrates robust detection of clinically relevant patterns revealed through heatmaps, including nocturnal wandering, counting delusional incidents, behavioral disturbances, quantifying silence and TV usage, respiratory irregularities, and nutritional adherence. The proposed framework establishes a scalable, privacy-first paradigm for continuous dementia monitoring, enabling earlier and more informed clinical interventions to push back hospitalizations and decompress elderly care facilities.