MemPalace

Best for: developers adding long-term memory to an AI application who want it running locally rather than through a memory-as-a-service API.

MemPalace is a local-first AI memory system built around verbatim storage with a pluggable backend. It reports 96.6% R@5 raw on the LongMemEval benchmark with zero API calls, and is distributed as a Python package with MCP support so agents can query it directly.

What makes it useful: the combination of local operation and published benchmark numbers. Memory systems are easy to claim and hard to evaluate, and MemPalace names the benchmark it scores against, which lets you compare it honestly against alternatives rather than taking marketing at face value.

Where it falls short: verbatim storage means the memory store grows with everything you feed it, so storage planning is your responsibility. It is one component of an application, not a finished product — you still build retrieval logic around it. The maintainers also warn about impostor sites: the only official sources are the GitHub repository, the PyPI package and mempalaceofficial.com. Other domains may distribute malware.

Details: Python, MIT licensed. Works with ChromaDB and MCP.