# SimpleMem

An efficient lifelong memory system for LLM agents that stores, compresses and retrieves long-term memories with semantic lossless compression, now covering text, image, audio and video.

SimpleMem is a memory system for LLM agents that stores, compresses and retrieves long-term memories using semantic lossless compression rather than keeping a raw, ever-growing transcript. It supports text memory through any AI platform that speaks the Model Context Protocol, and a Python integration for its full multimodal capabilities covering text, image, audio and video.

A single unified package auto-selects the text or multimodal backend based on the first method a caller uses, and includes an optional self-evolving retrieval component that diagnoses and improves its own retrieval strategies in a closed loop rather than using a fixed, hand-designed pipeline. It is described alongside benchmark comparisons against other memory systems on datasets such as LoCoMo and MemBench.

It installs as a Python package, ships an MCP server, and is licensed under MIT.

Website: https://github.com/aiming-lab/SimpleMem