Mem0

An open-source universal memory layer for AI agents, distributed under Apache 2.0. Provides persistent semantic memory backed by S3-compatible object storage with multi-signal retrieval combining semantic embeddings, BM25 keyword matching, and entity linking. The core differentiator is its **ADD-only extraction algorithm** — Mem0 never overwrites or deletes prior facts, instead appending new facts with temporal metadata so the agent can differentiate a user's past state from their present state. Repository: [github.com/mem0ai/mem0](https://github.com/mem0ai/mem0). Benchmark: published **LoCoMo score of 91.6** on long-context memory recall.

Standard vector databases lose chronological context — they treat retrieval as similarity over a flat semantic space, so an updated fact destroys the prior fact's existence in the index. Production agents need to reason about *when* something was true, not just *whether* it is true now. Mem0's ADD-only architecture preserves the chronological evolution of knowledge so an agent can answer "what did the user prefer six months ago" alongside "what does the user prefer now."

Persistent agent memory backed by S3, multi-turn conversational agents that survive across sessions, long-running automation that needs episodic state, RAG pipelines that benefit from temporal-aware retrieval.