What it is
Mem0 is a memory layer for AI apps and agents: it pulls the important facts out of each conversation (preferences, context, decisions) and brings them back when needed, so your assistant remembers each user without resending the whole history to the model. There's an open-source version that works with many vector stores and a managed platform with a dashboard and API.
How to use it
- Add memory to your agent in a few lines of code with the Python or TypeScript SDK.
- Store memories per user, session or agent and fetch them before each reply.
- Pick where they live: Qdrant, Chroma, Pinecone, pgvector and more in the open-source version.
- Connect facts with graph memory to capture relationships between people and topics.
- Review and edit what's remembered about each user from the platform dashboard.
What it's good for
- A WhatsApp sales assistant that remembers each customer's budget and preferences between chats.
- A wellness app whose virtual coach remembers every user's goals and progress.
- A support team that doesn't want its bot asking for details the customer already gave.
Pros
- Open source (Apache 2.0) with a managed option
- Cuts tokens sent to the model by retrieving only what's relevant
- Works with major vector databases and agent frameworks
- Separate memory scopes for users, sessions and agents
Cons
- Self-hosting also means running the vector store and the model that extracts memories
- Graph memory and higher limits sit on paid plans
- Dashboard and docs are English-only
