mem0

Mem0 is the self-improving memory layer for AI agents: it delivers personalized LLM experiences while cutting token costs by up to 80%.

Mem0 functions as a persistent memory engine for Large Language Models (LLMs), moving beyond simple vector storage. It uses LLM-based extraction to capture and store salient user preferences and factual data across sessions (e.g., 'User prefers aisle seats'). This intelligent data filtering cuts prompt tokens by up to 80% for developers, optimizing costs and latency. The system supports multiple memory types: semantic, episodic, and graph memory, mimicking human recall. Integration is seamless, requiring minimal code (three lines) and offering compatibility with major frameworks like OpenAI, LangGraph, and CrewAI.

https://mem0.ai/