Temporal knowledge graph and memory store for assistants.
A memory layer for AI chatbots that remembers conversation history and user facts — your chatbot never forgets important context.
Zep is a memory infrastructure platform for AI assistants that takes a fundamentally different approach from simple vector-based memory stores: it builds a temporal knowledge graph from conversations. Instead of storing flat memory facts, Zep extracts entities, relationships, and temporal information to create a structured graph that evolves over time.
The key insight behind Zep's architecture is that human conversations contain structured information — people mention other people, discuss events with dates, express changing preferences, and reference previous topics. Zep's extraction pipeline identifies these entities and relationships, creating graph nodes and edges that capture not just what was said but when, about whom, and how it relates to other known information.
In practice, this means Zep can answer queries that simple vector search cannot. 'What did the user say about Project X last month?' requires temporal filtering. 'Who did the user mention in connection with the budget issue?' requires relationship traversal. 'Has the user's opinion on remote work changed over time?' requires temporal comparison. These queries are natural in a knowledge graph but impossible with flat vector retrieval.
Zep offers both a cloud service and an open-source Community Edition. The cloud version provides a REST API, SDKs for Python and TypeScript, and a dashboard for exploring the knowledge graph. Integration with LangChain and other frameworks is available through compatible memory classes.
The platform also handles conversation summarization and message archival. Long conversations are automatically summarized to maintain relevant context without consuming excessive token budgets. Older messages are archived but remain searchable, creating a tiered memory system.
Zep's limitations are worth understanding. The knowledge graph extraction is computationally expensive — each message requires LLM processing to extract entities and relationships, which adds latency and cost. The quality of the graph depends on conversation richness; sparse or highly technical conversations may not produce useful graph structures. The Community Edition has limited features compared to the cloud version, and the temporal knowledge graph is primarily available in the commercial offering.
For applications where understanding temporal context and entity relationships matters — enterprise assistants, long-running customer relationships, complex project management — Zep's approach is genuinely more capable than flat memory stores. For simpler personalization use cases, it may be over-engineered.
Was this helpful?
Zep provides purpose-built memory infrastructure for AI assistants with strong temporal awareness and entity extraction. The dialog classification and structured data extraction features go beyond simple conversation storage. Users appreciate the focus on chat-oriented memory with automatic summarization. Limitations include a narrower scope than general-purpose vector databases and the fact that the cloud service is still maturing. Good choice for chat applications needing sophisticated memory management.
Extracts entities, relationships, and temporal information from conversations to build a knowledge graph that evolves over time. Supports temporal queries, relationship traversal, and change tracking across conversation history.
Use Case:
An enterprise assistant that tracks project stakeholders, their responsibilities, and how project status has changed across months of conversations.
Automatically identifies named entities (people, organizations, projects, dates) in conversations and links them to existing graph nodes. Resolves entity aliases and coreferences across messages.
Use Case:
Recognizing that 'John', 'John Smith', and 'the project lead' all refer to the same person across different conversations.
Automatically summarizes long conversations to maintain relevant context without consuming full token budgets. Summaries are hierarchical — recent messages are detailed while older context is progressively compressed.
Use Case:
Maintaining context from a 50-message customer support thread in a 500-token summary that captures all key decisions and open issues.
Active messages, summarized context, and archived history form a tiered system. Recent messages are available in full, older context is summarized, and archived messages remain searchable but aren't loaded by default.
Use Case:
An AI assistant that provides detailed context from today's conversation, summarized context from this week, and can search archived conversations from months ago when needed.
Extracts discrete facts from conversations and classifies them by type (preference, decision, goal, opinion). Facts are timestamped and linked to source messages for provenance tracking.
Use Case:
Extracting that a customer decided to upgrade their plan (decision), prefers email communication (preference), and wants to launch by Q2 (goal) from a sales conversation.
Organizes memory by user and session with separate knowledge graphs per user. Supports cross-session memory retrieval and user-level fact aggregation across all conversations.
Use Case:
Retrieving all known facts about a customer across their 20 previous support conversations before starting a new interaction.
Free
forever
Free
month
Ready to get started with Zep?
View Pricing Options →Enterprise AI assistants managing long-running customer or employee relationships where temporal context and entity tracking matter
Customer support agents that need to understand a customer's full history including who they've spoken with, what was decided, and how issues evolved
Project management assistants that track stakeholders, decisions, deadlines, and changing requirements across months of conversations
Applications where users need to query their AI's memory temporally — 'what did we discuss about X last month?'
Zep works with these platforms and services:
We believe in transparent reviews. Here's what Zep doesn't handle well:
Mem0 stores flat memory facts with vector retrieval. Zep builds a temporal knowledge graph with entity relationships and time-based queries. Zep is more powerful for complex queries but more expensive to operate. Mem0 is simpler and cheaper for basic personalization use cases.
Each message requires LLM processing for entity and relationship extraction, typically adding 1-3 seconds of latency. This can be processed asynchronously (message is handled first, graph is updated in background) to avoid blocking the user experience.
The Community Edition includes basic memory and summarization features. The temporal knowledge graph, advanced entity extraction, and relationship querying are primarily available in the commercial cloud offering. The open-source version is more comparable to simpler memory stores.
Yes. Zep provides a LangChain-compatible memory class (ZepMemory) that drops into existing chains. For LlamaIndex, there's a compatible chat store. The REST API also allows custom integration with any framework.
Weekly insights on the latest AI tools, features, and trends delivered to your inbox.
People who use this tool also find these helpful
Open-source vector database designed for AI applications with fast similarity search, multi-modal embeddings, and serverless cloud infrastructure for RAG systems and semantic search.
Open-source framework that builds knowledge graphs from your data so AI systems can reason over connected information rather than isolated text chunks.
Open-source embedded vector database built on Lance columnar format for multimodal AI applications.
LangChain memory primitives for long-horizon agent workflows.
Stateful agent platform inspired by persistent memory architectures.
Universal memory layer for AI agents and LLM applications. Self-improving memory system that personalizes AI interactions and reduces costs.
See how Zep compares to CrewAI and other alternatives
View Full Comparison →AI Agent Builders
CrewAI is an open-source Python framework for orchestrating autonomous AI agents that collaborate as a team to accomplish complex tasks. You define agents with specific roles, goals, and tools, then organize them into crews with defined workflows. Agents can delegate work to each other, share context, and execute multi-step processes like market research, content creation, or data analysis. CrewAI supports sequential and parallel task execution, integrates with popular LLMs, and provides memory systems for agent learning. It's one of the most popular multi-agent frameworks with a large community and extensive documentation.
Agent Frameworks
Open-source multi-agent framework from Microsoft Research with asynchronous architecture, AutoGen Studio GUI, and OpenTelemetry observability. Now part of the unified Microsoft Agent Framework alongside Semantic Kernel.
AI Agent Builders
Graph-based stateful orchestration runtime for agent loops.
AI Agent Builders
SDK for building AI agents with planners, memory, and connectors. - Enhanced AI-powered platform providing advanced capabilities for modern development and business workflows. Features comprehensive tooling, integrations, and scalable architecture designed for professional teams and enterprise environments.
AI Memory & Search
Universal memory layer for AI agents and LLM applications. Self-improving memory system that personalizes AI interactions and reduces costs.
No reviews yet. Be the first to share your experience!
Get started with Zep and see if it's the right fit for your needs.
Get Started →Take our 60-second quiz to get personalized tool recommendations
Find Your Perfect AI Stack →Explore 20 ready-to-deploy AI agent templates for sales, support, dev, research, and operations.
Browse Agent Templates →