Introduction

Fresh

Mem0 is a universal, self-improving memory layer for LLM applications that enables persistent context across sessions. It provides intelligent memory management that captures, organizes, and retrieves information automatically.

What is Mem0?

Mem0 solves a fundamental problem with LLM applications: they forget everything between sessions. Every conversation starts from scratch, forcing users to repeat context and preferences.

Mem0 adds a persistent memory layer that:

• Captures key facts, preferences, and decisions from conversations
• Organizes memories with smart deduplication and categorization
• Retrieves relevant context in sub-50ms for real-time applications
• Updates automatically when information changes or conflicts arise

Two Ways to Use Mem0

Mem0 Platform (Managed)

Production-ready in 5 minutes. No infrastructure to manage.

• Hosted vector store, graph services, and rerankers
• SOC 2 Type II compliant, GDPR ready
• Dashboard for visual memory inspection
• Auto-scaling and high availability

Get Started with Platform

Mem0 Open Source

Self-host for full control over data and deployment.

• Apache 2.0 license
• Choose your own LLM, vector DB, and embedder
• Offline capability for privacy-critical deployments
• Fully customizable codebase

Get Started with Open Source

How It Works

graph LR
    A[User Message] --> B[Mem0 Engine]
    B --> C[Extract Facts]
    C --> D[Check Conflicts]
    D --> E[Store Memory]
    E --> F[Vector Store]
    E --> G[Graph Store]
    H[Search Query] --> B
    B --> I[Retrieve Relevant Memories]
    F --> I
    G --> I

1. Messages come in -- User/assistant conversation turns are sent to add()
2. Facts are extracted -- LLM identifies key preferences, decisions, and entities
3. Conflicts are resolved -- Existing memories are checked for duplicates and contradictions
4. Memories are stored -- Vector embeddings and optional graph relations are created
5. Context is retrieved -- Semantic search returns relevant memories in sub-50ms

Key Features

Feature	Description
Smart Deduplication	Automatically detects and merges duplicate information
Conflict Resolution	Updates memories when new information contradicts existing ones
Graph Memory	Tracks entity relationships for richer context
Multimodal	Extracts memories from text, images, and documents
Custom Categories	Replace default tags with your own terminology
Webhooks	Real-time notifications for memory events
Memory Export	Structured export of all stored memories
Expiration	Automatic cleanup of time-sensitive memories

Platform vs Open Source

Feature	Platform	Open Source
Time to first memory	5 minutes	15-30 minutes
Infrastructure	Fully managed	Self-hosted
Vector DB	Managed	27+ options
LLM providers	Managed (optimized)	20+ options
Graph Memory	Managed	Self-configured
Webhooks	Built-in	Not available
Dashboard	Web-based	Not available
License	Usage-based	Apache 2.0 (free)
Best for	Fast iteration, production	Full control, cost optimization

Next Steps

• Platform Quickstart -- Get your first memory in 5 minutes
• Open Source Quickstart -- Self-host with Python
• Memory Types -- Understand the memory architecture
• API Reference -- Full REST API documentation