PMB is an open-source project memory system designed to give AI coding agents persistent context across development sessions. It helps developers avoid repeatedly explaining project details, decisions, and background information every time they start a new conversation with an AI assistant.
The platform connects with AI coding tools such as Claude Code, Cursor, Codex, and Zed through MCP, allowing agents to access a shared memory layer containing important project information. This includes architectural decisions, lessons learned, goals, documentation, recent changes, and other project-specific knowledge.
Unlike cloud-based memory solutions, PMB follows an offline-first approach. All data is stored locally in a SQLite workspace on the user's machine, with no cloud dependency, no API keys required, and no additional LLM calls needed when retrieving stored context. This keeps project knowledge private while reducing unnecessary costs and latency.
PMB also provides tools for managing and understanding stored memories. Users can inspect, export, and organize their project knowledge through a local dashboard, while impact tracking helps measure which memories actually improve AI-assisted development workflows.
By creating a persistent knowledge layer between developers and coding agents, PMB aims to make AI collaboration more consistent and efficient. It turns AI assistants from short-term chat tools into longer-term development partners that understand the history and context of a project.