Mempalace

MemPalace is a local first memory system that gives AI assistants long term recall of past conversations. It stores content verbatim and builds a compact index over it, so a model can find the exact original wording instead of working from a lossy summary. The design borrows from the method of loci. Memory is organized into wings for people, projects and topics, rooms for days or sessions, closets that group related material by thread, and drawers that each hold one untouched chunk of text. A compressed symbolic index written in a dialect the project calls AAAK lets a model scan thousands of entries in one pass and open only the drawer it needs. The index is for finding, while the content stays exactly as it was said. The vendor states that it runs entirely on the local machine, using ChromaDB on disk for retrieval and SQLite for a knowledge graph, with nothing uploaded or synced and no cloud service or API keys needed. The site refers to benchmark results and MCP tooling, links to documentation and the GitHub repository, and offers a mailing list for updates. It is aimed at people who want an AI assistant to remember personal and project details across sessions while keeping the data on their own hardware. The site publishes no pricing page or plan prices.