Memoir – Git for AI agent memory, with a Claude Code plugin (github.com)

Memoir has introduced a high-performance semantic memory system for AI agents, incorporating Git-like version control to address longstanding challenges in AI memory management. By replacing opaque vector databases with a transparent, versioned, and cryptographically secure memory storage, Memoir aims to tackle issues such as context contamination, token rent, and memory drift. This innovation allows AI agents to maintain a more organized and efficient memory structure, crucial for applications that require consistent and reliable performance. Significantly, Memoir enables the creation of hierarchical semantic paths for memory storage, allowing for clearer recall and streamlined operations. Users can branch, commit, merge, and rollback memories with ease, enhancing the agent's ability to revert to previous states without losing the entire memory store. This version control approach also supports faster lookups, automatic memory aggregation, and a clean architecture that separates storage, classification, and search layers. By integrating with Claude Code, Memoir simplifies the user experience with automatic context injections and memory snapshots, marking a notable advancement in how AI agents manage and utilize memory information. As an open-source project still in alpha, Memoir invites contributions from the community, particularly those focused on coding agents.