smrti

Smrti functions as a sophisticated memory engine, inspired by the AtomSpace model, designed to capture and manage an agent's experiences and knowledge. Its core features include: * Each memory node (atom) stores Bayesian truth value * Atoms also carry an emotional valence and attention weight * Retrieval prioritizes salience across five dimensions, not recency * Evidence is append-only for immutable observation logs * Multi-tenant isolation for distinct agent memories This system allows agents to remember what is contextually important rather than simply what occurred most recently. Truth values are dynamically consolidated during epochs using Probabilistic Logic Networks, ensuring beliefs evolve with new evidence without directly mutating past observations. The architecture integrates a six-layer system within a single SQLite file, providing vector indexing via `sqlite-vec` for efficient search and entity resolution through a cascade of matching techniques. Smrti's API offers six primary operations: `remember`, `recall`, `believe`, `reflect`, `forget`, and `personality`. This Python API is consistently exposed across various servers, enabling seamless memory interaction. It's built for developers and researchers creating autonomous agents where robust, context-aware memory and belief systems are critical for decision-making and learning.