Memind
A Java-native, self-evolving memory and context engine for AI agents that distills raw conversations, tool calls and documents into memory graphs and evolving insight trees, retrieved through REST, MCP and SDKs.
Memind turns raw context, conversations, tool calls, documents, images, audio and agent timelines, into structured memory and reusable experience, continuously organizing it into memory graphs, threads and an evolving Insight Tree that distills raw memories into Leaf, Branch and Root levels of understanding rather than storing isolated facts. It separates USER memory, profiles, preferences, life context, from AGENT memory, directives, tool experience, playbooks and resolved-task knowledge, so the same system can serve coding agents, local harness agents, chatbots, companions, copilots and workflow agents.
Retrieval spans Insight Trees, memory items, raw source data, memory graphs, memory threads, vector search and BM25 keyword search, temporal signals, and an optional deep-retrieval mode with query expansion, sufficiency checking and reranking. It is retrieved through REST, MCP, SDKs and first-party plugins for popular agents, and is built natively in Java rather than as a port of a Python memory system.
The project reports leading results, under an evaluation protocol aligned with MemOS and EverMemOS-style evaluation, across the LoCoMo, LongMemEval and PersonaMem long-memory benchmarks.