A-MEM Agentic Memory
Establishes a Zettelkasten-inspired agentic memory system for LLM agents, offering deep integration with IDE environments.
About
A-MEM is an advanced agentic memory system designed for LLM agents, built upon the principles of Zettelkasten. It offers seamless integration with popular IDEs like Cursor and VSCode via the MCP protocol. The system features a robust dual-storage architecture, combining ChromaDB for semantic retrieval with a configurable graph backend (NetworkX, RustworkX, or FalkorDB) for explicit, typed memory linking and graph traversal. Key functionalities include automatic note construction, link generation, memory evolution, and a suite of 'Memory Enzymes' that autonomously maintain graph health. Additionally, it incorporates a Researcher Agent for deep web research and context optimization, making it an invaluable tool for enhancing agent intelligence and knowledge management.
Key Features
- Dual-Storage: ChromaDB for vector search and graph backend (NetworkX/RustworkX/FalkorDB) for explicit typed edges
- Memory Enzymes: Autonomous background processes for graph maintenance (Link Pruner, Relation Suggester, Summary Digester)
- Researcher Agent: Deep web research for low-confidence queries with JIT context optimization and automatic note creation
- IDE Integration: Direct memory system access and management within Cursor and VSCode via MCP protocol
- Priority Scoring: Dynamic ranking of search results based on note type, age, usage, and graph connections
- 1 GitHub stars
Use Cases
- Empowering LLM agents with structured, persistent, and evolvable memory for enhanced performance and recall.
- Integrating advanced agentic memory and web research capabilities directly into developer IDEs (Cursor, VSCode).
- Automating the collection, organization, and semantic linking of research findings from the web and PDFs for agents.
