ReMe is a memory management kit for AI agents, built around a simple idea: memory should be stored as readable files that both agents and people can inspect and edit. Short for Remember Me, Refine Me, it turns conversations and resources into an interconnected Markdown knowledge base that agents can search and evolve over time.
ReMe is part of the AgentScope ecosystem and is open source on GitHub with about 3,660 stars, under active development in 2026. Durable memory lives in ordinary Markdown files with frontmatter and wikilinks, so it can be synced, backed up and edited with familiar tools while indexes remain rebuildable. Retrieval combines BM25, optional embeddings and wikilink expansion to surface precise, line-level passages together with their relationships, without stuffing the whole knowledge base into the context window. One workspace can be shared across personal assistants, coding agents and other agent frameworks, with plugins available for ecosystems such as DeepSeek Harness and OpenClaw. Because memory is versionable text, teams can diff what an agent has learned, roll back bad memories and share curated knowledge bases between projects, something that is effectively impossible with opaque embedding stores.
ReMe is free and open source under the Apache-2.0 license. It runs locally or on your own servers, with costs limited to any embedding or LLM APIs you choose to enable.
Agent developers who want transparent, file-based long-term memory that users can audit and edit, instead of an opaque vector database locked inside one framework.