Mem0

Mem0 ("mem-zero") is a memory layer for AI assistants and agents. It stores facts drawn from conversations and retrieves the relevant ones at inference time, so an application can remember user preferences and history across sessions. Memory is scoped at the user, session, and agent levels, and typical uses include customer support, personal assistants, healthcare, and productivity or gaming applications.

You can run it as a Python or Node.js library, as a self-hosted server with Docker Compose, a dashboard, and API keys, or on the hosted Mem0 Platform. An LLM is required; the default is gpt-5-mini with text-embedding-3-small for embeddings, and other providers are configurable. A mem0 CLI manages memories from the terminal and can mint an API key for an agent without an email or dashboard, with the human owner claiming the account later.

The memory algorithm introduced in April 2026 uses single-pass, add-only extraction (memories accumulate rather than being overwritten), stores agent-confirmed actions with the same weight as user facts, links extracted entities across memories, fuses semantic, BM25 keyword, and entity-match retrieval, and ranks dated memories with temporal reasoning. Reported benchmark results, measured on the managed platform in single-pass retrieval at a top-200 budget, are below; the evaluation framework is open source.

Benchmark	Previous	New	Tokens	Latency p50
LoCoMo	71.4	92.5	7.0K	0.88s
LongMemEval	67.8	94.4	6.8K	1.09s
BEAM (1M)	—	64.1	6.7K	1.00s
BEAM (10M)	—	48.6	6.9K	1.05s

Features

• Multi-level memory: user, session, and agent state retained and retrieved per scope
• Add and search API: memory.add(messages, userid=...) extracts facts; memory.search(query, filters=..., topk=...) returns the relevant ones
• Hybrid retrieval: semantic, BM25 keyword, and entity matching scored in parallel; the mem0ai[nlp] extra adds spaCy-based entity extraction
• Deployment options: library, self-hosted server with auth on by default, or the hosted platform, with a migration guide between them
• Agent signup: mem0 init --agent issues a working API key to an agent in a few commands
• Agent skills: installable skills for Claude Code, Codex, Cursor, Windsurf, OpenCode, and OpenClaw, including /mem0-integrate and /mem0-test-integration pipelines
• Framework integrations: LangGraph and CrewAI guides, a ChatGPT-with-memory demo, and a browser extension that stores memories across ChatGPT, Perplexity, and Claude
• Model flexibility: configurable LLM and embedding providers beyond the OpenAI defaults