Vibe Coding DiscoverSimpleMem
View on GitHub[ICML'26] SimpleMem: Efficient Lifelong Memory for LLM Agents — Text & Multimodal
SimpleMem is a lifelong memory stack for LLM agents: it stores dialogues and multimodal inputs as compressed atomic memories with embeddings, then retrieves them semantically. Ships a Python package, MCP server, cross-session memory, and a self-evolving retrieval tuner. MIT, Python.
Use Cases
Give LLM agents lifelong long-term memoryPersist and recall cross-session conversation contextStore and retrieve multimodal memories (text, image, audio, video)Semantic (lossless-compressed) memory retrieval to cut token usageExpose memory as an MCP server for Claude Desktop, Cursor, LM StudioAuto-tune retrieval hyperparameters via self-evolution loopBuild RAG knowledge pipelines over dialogue and mediaReproduce LoCoMo / MemBench memory benchmarks
Built With
- Language
- Python
- Frameworks
- LangChain · LangGraph · langmem · LiteLLM · FastAPI · LanceDB · Qdrant · sentence-transformers · PyTorch · HuggingFace Transformers · SQLite · Docker
Tags
long-term-memory · llm-agents · memory · rag · multimodal · semantic-search · retrieval · embedding · compression · knowledge-graph · mcp · cross-session-memory · audio · video · vision · self-evolving