LycheeMemory
LycheeMemory is a compact long-term memory framework for LLM agents that builds structured, consolidated conversational memory and retrieves it adaptively, distributed as an OpenClaw plugin, a Claude Code MCP integration, a Hermes plugin, and a Python API or HTTP MCP server.
LycheeMemory is a memory framework for LLM agents that starts from efficient conversational memory, built through structured organization, lightweight consolidation and adaptive retrieval, and extends toward action-aware, usage-aware memory for more capable agents. Semantic memory is stored in SQLite and LanceDB rather than a separate graph database, and a transformer-based reranker improves evidence selection during search.
It integrates with several agent runtimes: a native plugin for OpenClaw, an MCP-and-hooks integration for Claude Code, a runtime plugin for Hermes, a PyPI package for direct Python use, and an HTTP MCP server for any other MCP client. It also exposes OpenAI-compatible chat completion endpoints with request-level control over when memory is consolidated, and includes a visual memory module for multimodal content.