MemPalace/mempalace

What it solves

MemPalace provides a local-first AI memory system that allows users and AI agents to store and retrieve conversation history and project data verbatim. It solves the problem of "forgetting" in long-term AI interactions by avoiding summaries or paraphrasing, instead using semantic search to retrieve exact original content.

How it works

The system organizes data into a structured hierarchy: people and projects are "wings," topics are "rooms," and original content is stored in "drawers." It uses a pluggable retrieval layer (defaulting to ChromaDB) to perform semantic searches across this structured index. It can be integrated into AI workflows via a Model Context Protocol (MCP) server, a CLI, or auto-save hooks for tools like Claude Code and Cursor IDE.

Who it’s for

It is designed for developers and AI power users who want a private, local memory for their AI agents, as well as those using coding assistants (like Claude Code or Cursor) who need to retain long-term context across sessions without relying on cloud-based summaries.

Highlights

• Local-First Privacy: Data stays on the local machine unless the user explicitly opts in to external services.
• High Retrieval Accuracy: Achieves 96.6% R@5 raw retrieval recall on LongMemEval without requiring API calls or LLMs.
• Pluggable Backends: Supports multiple storage options including ChromaDB, SQLite, Milvus, Qdrant, and pgvector.
• Structured Memory: Uses a wing/room/drawer organization rather than a flat corpus for better scoped searching.
• MCP Integration: Includes 45 MCP tools for managing memory, knowledge graphs, and agent coordination.