aiming-lab/SimpleMem

SimpleMem: Efficient Lifelong Memory for LLM Agents — Text & Multimodal

What it solves

SimpleMem addresses the inefficiency of long-term memory for LLM agents. Traditional systems either store raw, redundant interaction history—which consumes excessive tokens—or use slow, expensive reasoning loops to filter noise. SimpleMem provides a way to store and retrieve memories with high information density, allowing agents to recall more information while using significantly fewer tokens.

How it works

SimpleMem operates as a unified memory stack consisting of three integrated components:

  • SimpleMem (Text Core): Uses a three-stage pipeline (Semantic Structured Compression, Online Semantic Synthesis, and Intent-Aware Retrieval Planning) to distill unstructured dialogues into compact, self-contained facts with resolved coreferences and absolute timestamps.
  • Omni-SimpleMem (Multimodal): Extends memory to text, images, audio, and video using entropy-driven selective ingestion, progressive retrieval (hybrid FAISS + BM25), and knowledge graph augmentation for cross-modal reasoning.
  • EvolveMem (Self-Evolution): An autonomous research loop (Evaluate $\rightarrow$ Diagnose $\rightarrow$ Propose $\rightarrow$ Guard) that continuously tunes retrieval hyperparameters and discovers new retrieval dimensions (like query decomposition) to improve performance over time.

Who it’s for

It is designed for developers building LLM-based agents that require long-term, cross-session memory across multiple modalities (text, image, audio, video) and those who want to optimize retrieval performance without manual hyperparameter tuning.

Highlights

  • Multimodal Support: Handles text, images, audio, and video memory.
  • Semantic Lossless Compression: Reduces inference-time token consumption by approximately 30x compared to raw history.
  • Self-Evolving Retrieval: Automatically optimizes its own retrieval configuration through an LLM-driven closed-loop process.
  • Broad Integration: Works via Python integration or as an MCP (Model Context Protocol) server for clients like Claude Desktop and Cursor.

Related

  • Project
    zjunlp/LightMem

    A lightweight memory management framework for LLMs and AI agents that provides efficient storage, retrieval, and update mechanisms for long-term memory.

  • Project
    LycheeMem/LycheeMem

    A lightweight long-term memory framework for LLM agents that uses structured semantic, procedural, and visual memory stores to maintain persistent context across sessions.

  • Project
    TeleAI-UAGI/telemem

    An agent memory management layer that provides long-term, character-aware text and video memory for AI agents, serving as a high-performance alternative to Mem0.

  • Project
    mnemon-dev/mnemon

    A persistent, LLM-supervised memory layer for AI agents that enables cross-session knowledge retention across multiple AI frameworks using a four-graph knowledge store.

  • Project
    MemoriLabs/Memori

    A persistent, structured memory layer for AI agents that captures conversations and and agent execution traces to eliminate session amnesia and reduce token costs.