Mem0

Mem0 gives AI applications persistent memory. Rather than replaying an entire conversation history on every call, it extracts the facts worth keeping, stores them, and retrieves only what is relevant to the current turn.

How it works

Mem0 processes conversations to identify durable information — preferences, decisions, relationships, project details — and maintains it as a memory store that is updated rather than appended. Contradictions resolve to the newer fact instead of accumulating.

Why that matters

• Cost and latency. Retrieving a handful of relevant memories is far cheaper than sending a long history every time.
• Quality. Long contexts degrade attention; targeted retrieval does not.
• Continuity. Users notice when an assistant forgets what they said last week, and they notice when it does not.

Features

• Multi-level memory scoped to user, session, or agent
• Graph memory for relationships between entities
• Any model and any vector store underneath
• Open-source SDK for Python and TypeScript, plus a managed platform with a dashboard and search

Free tier on the platform with usage-based pricing; the core library is open source. Widely used in companions, support agents, and personal assistants.