Show HN: Your self, in every light - a local-first MCP self model for AI agents (github.com)

Alma has introduced a local-first memory control point (MCP) server designed to enhance AI agents' understanding of individual users while prioritizing user privacy. By storing personal facts and preferences—such as names, roles, and working styles—locally, Alma enables agents to access a scoped version of this information without requiring permanent write access. This prevents the fragmentation of user memory across various AI tools, where each application learns a different version of the user, and mitigates the risks of untrusted agents altering user memory invisibly. The key technical features of Alma include the use of "facets" as organized facts, with each facet associated with a specific context and one-time approval tokens needed for durable writes. The system is designed for transparency and control, with all changes auditable through an append-only event log. Agents can only read from the memory as permitted and can propose new facets for user approval, ensuring that users maintain authority over their own data. Essentially, Alma empowers users by allowing AI systems to adapt to them without compromising their privacy or control, providing a significant step forward in user-centric AI development.