MemPalace

mempalace

AI🌱 NEW PROJECT BOOST#RAG#Vector Database#LLM#Knowledge Graph#Python

// summary

MemPalace is a local-first AI memory system that stores conversation history as verbatim text for high-accuracy semantic retrieval. It utilizes a structured indexing approach with pluggable backends to organize content into wings, rooms, and drawers without requiring external API calls. The platform also features a temporal knowledge graph, MCP tools, and agent-specific diaries to provide comprehensive context management.

// technical analysis

MemPalace is a local-first AI memory system designed to store conversation history and project data as verbatim text, avoiding the information loss associated with summarization or paraphrasing. It utilizes a structured storage architecture—categorizing data into wings, rooms, and drawers—to enable scoped semantic search rather than flat corpus scanning. By providing a pluggable retrieval backend and operating entirely offline, the project prioritizes data privacy and high-performance retrieval, achieving a 96.6% R@5 recall on LongMemEval without requiring any API calls.

// key highlights

Verbatim storage ensures that conversation history is preserved exactly as-is without lossy summarization or extraction.

A hierarchical storage structure organizes data into wings, rooms, and drawers to allow for precise, scoped semantic searching.

The system features a pluggable retrieval backend, allowing users to swap storage engines like ChromaDB without modifying the core system.

It achieves high retrieval recall (96.6% R@5) using purely local semantic search, eliminating the need for external API calls or cloud dependencies.

Includes a temporal entity-relationship knowledge graph backed by SQLite to manage complex data associations and validity windows.

Provides 29 MCP-compatible tools to facilitate seamless integration with agents, cross-wing navigation, and automated diary management.

// use cases

Verbatim conversation history storage and semantic retrieval

Structured knowledge graph management with temporal entity-relationship tracking

Integration with AI agents and Claude Code via MCP tools and auto-save hooks

// getting started

To begin, install the package via pip and initialize your palace directory using the CLI command 'mempalace init'. You can then populate your memory by running 'mempalace mine' on your project files or conversation logs. Finally, use 'mempalace search' to query your data or 'mempalace wake-up' to load context for new sessions.