Phloem – Local-first AI memory with causal graphs (MCP server, zero network) (github.com)

Phloem, a new local MCP server, has been introduced to provide AI tools with persistent memory across sessions, addressing a significant limitation where AI models lose context upon tab closure. This innovative solution integrates with any tool that supports the Model Context Protocol (MCP), including popular platforms like Claude Code, VS Code, and Zed. Users can easily set it up, allowing all compatible AI tools on their machine to leverage a unified memory system without network dependence, as memories are stored in a local SQLite database. The significance of Phloem lies in its ability to create dynamic, causally linked memories, enhancing AI understanding and communication. Unlike static text files, Phloem memories are generated naturally during interactions and can be retrieved through semantic search, enabling AI to understand relationships between decisions and code changes. For example, if a JWT token expiration is altered, Phloem tracks this change and adjusts its memory confidence accordingly. This robust memory system not only facilitates better context switching between tools but also captures the underlying reasons for code decisions, thus making AI interactions more intelligent and context-aware.