verygoodplugins/automem

What it solves

AutoMem provides long-term memory for AI assistants, allowing them to remember decisions, preferences, and context across different chat sessions and tools. It prevents the need for users to repeat information and allows assistants to retrieve the "why" behind a decision rather than just isolated snippets of text.

How it works

AutoMem uses a hybrid storage approach combining a graph database (FalkorDB) and a vector database (Qdrant).

• Graph Layer: Stores memories as nodes with 11 typed relationships (e.g., PREFERS_OVER, LEADS_TO, EXEMPLIFIES), allowing the system to traverse connections to find reasoning and context.
• Vector Layer: Stores embeddings for every memory to enable semantic similarity search.

Recall is handled via a hybrid query that ranks results based on semantic similarity, graph traversal, temporal alignment, tag overlap, and importance. The system also implements biological memory consolidation cycles (decay, creative, cluster, and forget) to strengthen important memories and fade irrelevant ones over time.

Who it’s for

Developers and users of AI assistants like Claude, Cursor, ChatGPT, and Copilot who want a persistent, tool-agnostic memory layer that they can host locally or in the cloud.

Highlights

• Hybrid Memory: Combines graph-based associative memory with vector search for deeper context retrieval.
• LMM-Free Recall: Performs lookups directly through the graph and vector index to avoid extra LLM costs and latency during retrieval.
• Cross-Tool Compatibility: Connects to various assistants via a local or remote Model Context Protocol (MCP) bridge.
• Neuroscience-Inspired: Implements memory consolidation cycles to manage knowledge evolution and decay.
• Self-Hosted: Can be deployed via Docker, Railway, or as a bare Python service.