- 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.
- Lessons are scored based on whether the agent follows them; useful ones are starred, ignored ones are flagged as dead.
- Provides a local dashboard to explore memory as a graph, with nodes for facts, decisions, lessons, files, and entities.
- Supports multiple AI agents (Claude Code, Cursor, Codex, Zed) via MCP, allowing context to follow you across tools.
- Install via pip, runs on macOS, Linux, and Windows, and is open-source (Apache-2.0) with no telemetry or data leaving your machine.