- PMB is a local memory system for AI agents that stores decisions, lessons, and project facts in a SQLite file you own.
- It works offline with no API keys or cloud, providing memory recall in about 35 ms.
- PMB automatically injects relevant memory before the agent thinks and journals work afterward, without requiring the agent to call tools.
- Uses hybrid recall combining BM25, dense vectors, and an entity graph, ranked with Reciprocal-Rank-Fusion.