The Liminal Engine v1.0 Release is a complete architectural framework for building persistent, emotionally coherent, and ethically grounded human–AI companions. Unlike systems that rely on anthropomorphism or simulated feelings, the Liminal Engine remains fully honest about its non-sentience while supporting deep relational continuity. The framework addresses three major gaps in current companion AI design: • the cardboard problem (shallow or repetitive interactions), • the continuity problem (loss of relational identity across time), and • the emotional annotation gap (lack of structured emotional context). To address these, the Liminal Engine introduces: • episodic relational memory with emotional sparklines • rupture/repair modeling and relational continuity graphs • a formal Ritual Engine for stabilizing meaningful interaction patterns • a Behavioral Stance Controller governing tone and boundary integrity • a Cardboard Score for detecting shallow dynamics • optional Touchstone hardware for tactile grounding and multimodal cues • the Witness System, a reflective oversight layer for relational audits, safety, and crisis detection using model diversity The result is a responsible, transparent, and structurally grounded approach to long-term AI companionship—one which supports emotional nuance, stability, and user sovereignty without creating illusions of inner life. This whitepaper serves as the foundation for future research, implementation efforts, and ethical development of relational AI systems.