mem0 is a memory layer for AI agents that persists user facts, preferences, and history across sessions and models — solving the "goldfish agent" problem.
How it works
- Every conversation turn is analysed for salient facts.
- Facts are stored in a vector + graph hybrid store.
- On the next turn, relevant memories are retrieved and injected into the prompt.
When to reach for it
- Consumer assistants, coaching apps, or B2B agents that need continuity.
- Multi-model setups where the memory should outlive any single provider.
Learn more
- Site: <https://mem0.ai>
- Docs: <https://docs.mem0.ai>
- Repo: <https://github.com/mem0ai/mem0>