Memento Offers Shared, Durable Memory for AI Agents
TL;DR. A new project called Memento provides a shared, durable memory system designed for multiple AI agents, using Git for version control. - It stores long-term knowledge beyond individual conversations, accessible to various agents like people and services. - The system uses Markdown files for concepts and Git for history, with an FTS5 and vector index for search. - A local Needle model classifies natural-language requests into specific read operations, ensuring controlled access.
- Memento implements a shared, durable memory system for multiple AI agents.
- It leverages Git for version-controlled concept storage and history management.
- A fine-tuned local model routes natural language queries to specific read operations, maintaining system integrity.
- The architecture supports review workflows for proposed changes before integration into the shared repository.
Sources
- Memento – Shared, durable memory for multiple AI agents, with MCP access — rcarmo.github.io