Long-term memory for AI coding agents as a tree of plain files (github.com)

A revolutionary long-term memory system for AI coding agents has been introduced, allowing agents like Claude Code, Codex, and opencode to retain key information across sessions and projects. This memory system enables agents to remember decisions, environment setups, preferences, and corrections, effectively eliminating redundancy in user interactions. With the ability to share a single memory across multiple agents, users can ensure consistency and save time as agents retrieve relevant information efficiently — typically within 200 to 900 tokens — by navigating a hierarchical tree structure instead of scanning through large datasets. The significance of this development lies in its local, private, and accessible design, with memories stored as plain text files on the user's disk, providing full control over data without relying on cloud solutions. The system also flags outdated information while preserving historical context, making it adaptable over time. Compatibility with existing AI agents, ease of installation, and the elimination of external dependencies emphasize a user-friendly approach for AI practitioners. This innovation sets a new standard for collaborative coding tools, making AI assistants significantly more effective in managing long-term knowledge and improving user experiences in programming tasks.