Mem0 iconMem0

oss Freemium Star58k

Universal memory layer for AI Agents that enables personalized, context-aware interactions

57K+ GitHub Stars
91.6 LoCoMo Score
Y Combinator S24 Batch

Overview

Mem0 ("mem-zero") is an intelligent memory layer that enhances AI assistants and agents with persistent, personalized memory. Rather than treating each conversation as a blank slate, Mem0 remembers user preferences, adapts to individual needs, and continuously learns over time. It supports multi-level memory (User, Session, and Agent state) with features like entity linking, temporal reasoning, and hybrid search combining semantic, keyword, and entity matching. Ideal for customer support chatbots, AI assistants, healthcare applications, and autonomous agent systems that need to maintain context across interactions.

The Verdict

Who Should Use Mem0?

Best For

  • Teams building personalized AI assistants
  • Customer support chatbots needing history
  • Autonomous agent systems
  • Healthcare AI requiring patient context
  • Developers wanting plug-and-play memory

Not Ideal For

  • Simple stateless chatbots
  • One-off query applications
  • Teams needing deep graph relationships (try Letta)
  • Projects requiring only session memory

What's Great

  • Best-in-class benchmark scores (91.6 LoCoMo, 94.8 LongMemEval)
  • Token-efficient single-pass memory extraction
  • Multi-level memory (User, Session, Agent)
  • Flexible deployment: library, self-hosted, or cloud
  • Strong OSS community
  • Y Combinator backed (S24)
  • Simple API with Python and Node.js SDKs
  • Entity linking and temporal reasoning built-in

Watch Out For

  • Requires LLM for memory operations (adds cost)
  • Cloud pricing can scale with usage
  • Self-hosted requires infrastructure management
  • Less mature than RAG-focused alternatives for pure retrieval
  • Learning curve for optimal memory schema design

Pricing

View all features & details

Core Features

  • Multi-level memory (User, Session, Agent)
  • Entity linking across memories
  • Temporal reasoning for time-aware retrieval
  • Hybrid search (semantic + BM25 + entity)
  • Single-pass ADD-only extraction
  • Agent-generated facts as first-class
  • Cross-platform SDKs (Python, Node.js)
  • CLI for terminal management

Deployment Options

  • pip install mem0ai (library)
  • Docker Compose (self-hosted)
  • Cloud Platform (managed)
  • CLI: npm install -g @mem0/cli

Integrations

  • OpenAI (default: gpt-5-mini)
  • Anthropic Claude
  • Multiple LLM providers
  • Vector stores (Qdrant, etc.)
  • Vercel AI SDK
  • Claude Code, Cursor, Windsurf skills

Use Cases

  • AI Assistants with context
  • Customer Support bots
  • Healthcare patient history
  • Productivity tools
  • Gaming environments
  • Autonomous agent systems

Benchmarks

91.6
LoCoMo
Long-context memory benchmark (+20 pts over previous algorithm)
94.8
LongMemEval
Long-term memory evaluation (+27 pts improvement)
64.1
BEAM (1M)
Production-scale memory at 1M tokens
~7K
Token Efficiency
Avg tokens per retrieval with sub-1s latency

Real-World Usage

Community Stats

  • 6,500+ forks
  • Y Combinator S24 company
  • Active Discord community

Agent Skills Support

  • Claude Code integration
  • Cursor, Windsurf support
  • Vercel AI SDK compatible
  • Agent signup in 5 seconds

How It Compares

Feature Mem0 GBrain MemClaw
GitHub Stars 57K+ 20.8K N/A (Commercial)
Memory Type Multi-level (User/Session/Agent) Self-wiring graph + synthesis Enterprise shared memory
Entity Linking Yes, built-in Yes, typed edges Yes, auto-extracted
Synthesis/Gap Analysis No Yes, cited answers No
Governance (RBAC/Audit) Basic None Built-in
Multi-Agent/Fleet No Multi-brain federation Yes
Self-Hosted Yes (Docker) Yes (local-first) No
Open Source Apache 2.0 MIT No
Best For Personalized AI assistants Knowledge synthesis Enterprise agent fleets

User Reviews

Loading reviews...