mem0

Mem0 is an open-source project that adds a persistent memory layer to AI assistants, agents, and applications. It aims to let AI systems remember user preferences, session context, and agent state, and to keep learning over time. The README presents it as suitable for customer-support chatbots, AI assistants, and autonomous systems. The project offers several ways to use it: - Python library (`pip install mem0ai`) and npm package (`mem0ai`) for prototyping and testing. - A self-hosted server started with Docker Compose, including a dashboard, auth/API keys, and a bootstrap setup. - A managed cloud platform (app.mem0.ai) with all advanced features. - A CLI (`@mem0/cli` or `mem0-cli`) for adding and searching memories from the terminal. The core API is simple: instantiate `Memory`, then search relevant memories and add new memories from conversation messages. Mem0 requires an LLM (OpenAI's gpt-5-mini by default, with many other LLMs supported) and an embedding model (default text-embedding-3-small). Optional NLP extras enable hybrid search with BM25 keyword matching and entity extraction. The README also describes a new memory algorithm that uses single-pass add-only extraction, agent-generated facts, entity linking, multi-signal retrieval (semantic, BM25, entity), and temporal reasoning. Benchmark tables in the README report strong results for the managed platform (e.g., 92.5 LoCoMo, 94.4 LongMemEval) but note that open-source SDK users should expect directionally similar rather than identical numbers. The evaluation framework is open-sourced. Other highlights: an agent signup flow that mints API keys via CLI, reference/pipeline skills for AI coding assistants, integrations with ChatGPT/Perplexity/Claude browser extension, LangGraph support, CrewAI integration, documentation, and an Apache 2.0 license.