Show HN: server-noonien – conflict-free agent memory across machines, no server (github.com)

A new tool called server-noonien has been announced, offering a shared, serverless, and conflict-free memory solution for AI agents. By utilizing peer-to-peer synchronization, this tool acts as a drop-in replacement for the traditional modelcontextprotocol’s server-memory, enabling multiple machines to access a unified knowledge graph without needing a central server or database. Each machine maintains its own append-only shard and communicates directly with others to replicate memory, significantly reducing the chances of data conflicts typically seen in shared memory solutions. This innovation is particularly significant for the AI and machine learning community as it addresses common pitfalls associated with local agent memory management, especially when multiple machines are involved. Unlike conventional servers that overwrite shared files, server-noonien ensures that operations are merged in a non-destructive manner, thanks to its CRDT (Conflict-free Replicated Data Type) architecture. This allows for seamless collaboration among agents distributed across different machines, making it an attractive alternative for developers looking for a robust, scalable solution that also works offline. With full local replicas and optional storage solutions, server-noonien empowers users to maintain control over their data while leveraging a decentralized infrastructure.