ReMe

Active
GitHub Python Apache-2.0

Description

ReMe: Memory Management Kit for Agents - Remember Me, Refine Me.

Key Features

  • File-based memory system (ReMeLight) using Markdown files for transparent, editable, and portable agent memory
  • Vector-based memory with hybrid retrieval combining semantic search vectors and BM25 keyword matching
  • Automatic context compaction that compresses long conversations while preserving key information
  • Long-term memory persistence via structured summarization with file read/write/edit tools
  • Session memory support for in-conversation state management and cross-session memory inheritance
  • Tool result compaction for managing long outputs from external tool calls within context limits

Use Cases

💡 Personal AI assistants requiring long-term memory of user preferences and conversation history
💡 Customer service bots tracking issue history and user preferences across sessions
💡 Coding assistants maintaining consistent development experience with project context and code style preferences
💡 Task automation systems learning success and failure patterns from historical execution data

Tags

agent-memory rag memoryscope python

Categories

🧩 Agent Memory

Quick Start

Clone the repo: git clone https://github.com/agentscope-ai/ReMe.git && cd ReMe. Install: pip install -e '.[light]'. Set environment variables LLM_API_KEY and LLM_BASE_URL for your LLM provider. Initialize ReMeLight with a working directory, call start() to begin, and use compact_memory() and memory_search() to manage agent memory across conversations.
Visit GitHub

Related Projects