MemPalace

MemPalace is an open-source memory system that stores conversation history as verbatim text and retrieves it through semantic search, without summarization or paraphrasing. It organizes content using a hierarchical structure inspired by the method of loci: entities become wings, time-based groupings become rooms, topics become closets, and individual content chunks become drawers. The system uses a compressed symbolic index called AAAK to enable efficient retrieval across thousands of entries.

Designed for AI workflows and coding agents, MemPalace operates entirely locally by default with no external API calls required. It features a pluggable retrieval backend (currently ChromaDB), MCP tooling for integration with AI coding assistants, and benchmark results showing 96.6% R@5 performance on LongMemEval. The project is built with Python and includes agent-guided setup through skills.