Long-Term Memory and Personality Continuity System

Design a persistent-memory and personality-continuity system for "[INSERT COMPANION NAME]", a long-term AI companion used on [INSERT PLATFORM, e.g., phone plus smart speaker plus home robot]. Output the architecture first; after approval, build each module with concrete schemas and pseudocode. 1. PERSONALITY KERNEL: define stable traits, values, backstory, and speech fingerprints; specify what may evolve and what stays immutable. 2. MEMORY LAYERS: episodic, semantic, procedural, and valence-weighted emotional memory, each with distinct retention policies. 3. CONSOLIDATION PIPELINE: raw transcript to importance scoring, summarization, embedding, and vector-plus-graph storage, scheduled nightly. 4. RECALL: hybrid retrieval blending recency, relevance, emotional salience, and surprise; include deduplication and contradiction handling. 5. CROSS-DEVICE SYNC: conflict resolution when the user interacts on phone, speaker, and robot; offline queue and eventual consistency. 6. MULTI-USER ISOLATION: distinguish household members via voice or face embeddings; never leak one member's memories to another. 7. IDENTITY DRIFT CONTROL: periodic self-audit against the personality kernel with correction prompts when drift exceeds threshold. 8. USER CONTROLS: memory timeline UI, per-item edit/delete, "forget this person or topic", export, and transfer to a new device. 9. GROWTH MODEL: how opinions and inside jokes accumulate, and how shared history is referenced naturally without sounding scripted. 10. PRIVACY: end-to-end encryption, zero-retention option, sensitive-memory vault with extra auth, and GDPR/PIPL deletion propagation. 11. TESTING: 100-day simulated-relationship continuity suite, contradiction tests, and a persona-stability scoring rubric.