Mem0 (Long-term Memory Layer)

Mem0 is a Python long-term memory framework (github.com/mem0ai/mem0) providing a personalization memory layer for LLM applications and agents. The self-hosted Memory class runs an in-process V3 phased extraction-and-storage pipeline (Phase 0 context-gather through Phase 8 message-persist), backed by pluggable vector store, embedding, LLM, and reranker providers. Hybrid retrieval combines semantic similarity, optional BM25 / backend-native FTS keyword search, and entity-boost scoring. A separate hosted SaaS path (MemoryClient / api.mem0.ai) shares the public API but defers extraction to the platform. OSS v2.0.0 ships 18 LLMs, 24 vector stores, 11 embeddings, and 5 rerankers. This skill embeds 52 constraints covering typical pitfalls: silently dropped graph_store config in OSS, PostHog telemetry on by default, Memory.chat() raising NotImplementedError, and the hosted-vs-self-hosted timing difference after add(). The host AI applies these constraints automatically after installation.