📚 Memtrace: Structural Memory for AI Coding Agents

In short, this project tries to fix the "amnesia" problem of AI coding assistants. Memtrace is a structural memory system designed for coding agents like Cursor and Claude Code, using a bi-temporal graph to track the evolution and history of a codebase. The interesting part is its architectural approach: it is MCP-native and builds its memory graph with zero LLM calls. This means it relies entirely on local static analysis and structured data to understand codebase context, rather than burning expensive tokens to have a model summarize things. When dealing with large repositories, AI assistants often lose context or forget the logic behind previous edits. Memtrace attempts to solve this by attaching a reliable, deterministic graph-based long-term memory to these agents.