GrayMatter

GrayMatter is a memory system for AI agents that persists facts across sessions and reduces context token usage by 90%. It works as a general-purpose MCP server compatible with Claude Code, Cursor, Codex, and other MCP clients, or as a plain Go library. The tool runs as a single ~10 MB binary with no Docker, databases, or cloud accounts required.

Key features include semantic recall that injects only the top-8 relevant facts instead of full conversation history, a self-building knowledge graph with typed entities and co-mention edges, and deterministic decay where facts fade over 30 days of inactivity. Agents can curate their own memories through the memory_reflect tool, and facts export to Obsidian. Installation takes under a minute via package managers or pre-built binaries.