smrti – Graph-based memory engine for AI agents

Summary: smrti is a graph database designed for AI agent memory, storing nodes with Bayesian truth values, emotional valence, and attention weights. It ranks retrieval by salience across multiple dimensions rather than timestamps, enabling agents to remember relevant information beyond simple similarity searches.

What it does

smrti stores memories as graph nodes representing concepts, beliefs, episodes, and goals, all within a single SQLite file. Retrieval combines vector search with graph traversal and a salience formula that factors in similarity, attention, confidence, and emotional intensity.

Who it's for

It is intended for developers building AI agents that require persistent, contextually relevant memory across sessions, especially where confidence and emotional context matter.

Why it matters

It addresses the limitation of agents forgetting between sessions by providing a memory system that prioritizes important, connected, and emotionally weighted information rather than relying solely on vector similarity.