Mem0ai

Mem0 solves the persistence problem for conversational AI agents — by default, LLM context windows reset per session, but Mem0 provides a managed store where agents write observations and retrieve relevant memories at the start of each turn. The API is clean: add a memory, search for relevant memories, delete outdated ones. Self-hostable (open-source) or Mem0 Cloud. Strong fit for assistants, customer-support agents, and any agent needing cross-session continuity. Confidence is docs-derived.

REST API and SDK (Python, TypeScript). Core operations: memory.add(messages, user_id=...) stores new memories from conversation turns. memory.search(query, user_id=...) retrieves semantically relevant memories. memory.get_all(user_id=...) lists all memories. memory.delete(memory_id) removes a specific memory. memory.update(memory_id, data) updates stored content. Memories are scoped by user_id, agent_id, or run_id. Vector-backed semantic search. Mem0 Cloud REST API mirrors SDK semantics.

API key auth: MEM0_API_KEY environment variable or passed to MemoryClient(api_key=...). Keys from Mem0 Platform dashboard. HTTPS enforced. No OAuth2. Self-hosted: no key required; local vector store. Memory scoping (user/agent/run) provides logical access control within an account.

SDK raises exceptions on API errors; network failures surface as connection errors. Self-hosted: memory persistence depends on configured vector store (Qdrant, Chroma, etc.) and database reliability. Mem0 Cloud uptime at status.mem0.ai. Memory relevance depends on embedding model quality; fallback to get_all is safe when search precision is uncertain.

docs.mem0.ai covers quickstart, Python/TS SDK reference, memory scoping, self-hosting, and integration guides with LangChain and CrewAI. Getting started: pip install mem0ai, set API key, first memory add/search in under 5 minutes. Free tier on Mem0 Cloud. Open-source (GitHub: mem0ai/mem0). Community via Mem0 Discord.