MemPalace is a local-first, verbatim AI memory system designed to preserve conversation context in AI coding sessions without summarization. It provides semantic search across structured memory with pluggable storage backends and reports 96.6% R@5 raw recall on LongMemEval without requiring LLM or API calls.
Project overview
The project addresses conversation history loss by storing content verbatim rather than summarizing it, and it reports 96.6% R@5 raw recall on LongMemEval without LLM or API calls.
Project type
MCP · AI Agent · RAG
Use cases
Knowledge Q&A · Coding & Development
Deployment
Refer to project documentation
License
MIT
Best for
Developers and AI engineers needing a local-first memory system to store and retrieve AI coding conversation context without summarization.
Key capabilities
Stores conversation history and content as verbatim text without summarization.
Provides scoped searches across structured memory wings, rooms, and drawers.
Allows swapping the retrieval backend (examples include ChromaDB, SQLite, and Milvus) via interface contract.
Exposes 36 MCP tools for palace reads/writes, graph operations, and agent management.
Provides a local SQLite-backed entity-relationship graph with validity windows.
Integrates hooks for Claude Code, Codex CLI, and Cursor IDE to save sessions periodically.
Limitations and risks
Achieving the highest retrieval accuracy via LLM rerank requires supplying an external capable model.
Getting started
Installation is rated as easy because the system installs via uv tool or pipx with one command. The documented first success path is to run 'uv tool install mempalace', followed by 'mempalace init ~/projects/myapp', and then 'mempalace mine ~/projects/myapp'.
Alternatives and comparisons
Provides AI agents and AI coding assistants with web crawling and RAG capabilities over crawled content, serving as an alternative for acquiring external knowledge context.
Transforms a codebase into a semantically searchable knowledge graph so AI agents can reason about code architecture, offering an alternative for structural code context.
An AI programming assistant and tutorial project that teaches LangChain4j usage while serving as a mentor for programming learning, providing a different approach to developer assistance.
GitHub project description: The best-benchmarked open-source AI memory system. And it's free.
README: Claude Code sessions expire in 30 days without auto-save hooks wired.
README: MemPalace stores your conversation history as verbatim text and retrieves it with semantic search. It does not summarize, extract, or paraphrase.
README: ChromaDB is the default and needs no configuration. MemPalace also ships a pluggable backend contract