MemPalace/mempalace
The best-benchmarked open-source AI memory system. And it's free.
What it solves
MemPalace provides a local-first AI memory system that stores conversation history as verbatim text rather than summaries. This prevents information loss from paraphrasing and allows for high-accuracy retrieval of specific details from past interactions, which is critical for maintaining long-term context in AI sessions.
How it works
The system uses a structured index where people and projects are organized into "wings," topics into "rooms," and original content is stored in "drawers." It employs semantic search to retrieve verbatim text, with a pluggable backend architecture that supports various vector databases (ChromaDB by default, but also Milvus, Qdrant, and pgvector).
Who it’s for
Developers and AI power users who want a private, local memory layer for their AI agents or IDEs (like Claude Code or Cursor), ensuring that their conversation history is preserved and verbatim retrieval is paljony
Highlights
- Local-first Privacy: Nothing leaves the machine unless the user opts in; no API keys are required for core functionality.
- High Retrieval Accuracy: Achieves 96.6% R@5 on LongMemEval without any LLM calls.
- Pluggable Backends: Supports multiple storage substrates including ChromaDB, Milvus, and Qdrant.
- MCP Server Integration: Includes 36 MCP tools for managing memory, knowledge graphs, and agent diaries.
- Auto-save Hooks: Provides specialized hooks for Claude Code, Codex CLI, and Cursor IDE to automate history preservation.
- Temporal Knowledge Graph: Includes a local SQLite-backed graph for tracking entity relationships with validity windows.