Mosaic: Local MCP server for agent memory (github.com)

Mosaic, a new local memory control protocol (MCP) server, has been announced, enabling structured memory storage for AI agents. It leverages HexxlaDB to maintain agent memory directly on user-operated infrastructure, with features like scoped context management, spatially-organized memory access, and conflict handling that enhances the reliability of AI interactions. This is significant for the AI/ML community as it addresses a critical issue where agents often struggle with drifting facts and disjointed sessions by ensuring a consistent and organized memory that evolves over time. Key technical components of Mosaic include a hex lattice structure for spatial memory organization, hybrid retrieval methods combining semantic similarity with structured constraints, and a seamless handling of conflicting information. Its YAML-based configuration allows users to define retention policies and control memory writes, providing a transparent method for managing agent behavior. The framework supports both encrypted and unencrypted database options, promoting user data sovereignty. Moreover, the open-source nature of Mosaic invites community collaboration and development, paving the way for future enhancements in agent memory capabilities and reliability.