nxm-memory is a local MCP server that indexes workspace files and makes them queryable in natural language without cloud transmission. It reduces token usage by compressing source code into structural maps and retrieving only relevant document chunks via semantic search, enabling AI assistants like Claude Code to fit more context within token limits while maintaining privacy.
ContextOS is a system designed to optimize AI coding agents by reducing context window saturation through spatial architecture visualization and AST-sliced task contexts. It provides a native macOS application with interactive architectural blocks and an MCP server that delivers precise symbol facades to agents, achieving 99.4% token reduction in real-world testing.