Personalized assistants
Carry durable user preferences, history, relationships, and current account context across many conversation threads.
Independent tool overview
Zep at a glance
Zep is context-engineering infrastructure for AI agents. It turns chat, business events, documents, and other text or JSON into temporal context graphs, then retrieves relevant, provenance-linked memory for prompts and agent tools.
Visit the official Zep site ↗
- Category
- Agent memory and context-engineering infrastructure
- Core model
- Temporal Context Graph built from chat, text, JSON, documents, and business events
- Developer access
- Cloud API and SDKs for Python, TypeScript, and Go; works with any agent framework or none
- Open-source option
- Graphiti framework under Apache 2.0; infrastructure and model costs are separate
- Starting price
- Free prototype allowance of 10,000 credits per month; paid Flex starts at $125/month
- Last reviewed
- September 1, 2026
Overview
What Zep is
Zep is built for developers and organizations that need an AI agent to remember more than the current conversation. Applications send user messages and business data to Zep, which extracts entities and time-aware facts into a Context Graph. The application can then request a ready-to-use context block or search the graph directly.
The main distinction is temporal memory: when new information supersedes an old fact, Zep preserves when each fact was valid instead of treating every statement as equally current. This is useful for assistants that need evolving customer preferences, account state, support history, transactions, decisions, or other changing relationships.
Zep Cloud is the commercial managed service, with self-serve and enterprise deployment options. Graphiti is Zep's separate Apache-2.0 open-source framework for building temporal knowledge graphs. Choosing Graphiti means operating the graph database, model integrations, indexing, scaling, and governance yourself; it is not a free self-hosted edition of the current managed Zep platform.
Use cases
Who Zep is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Customer and support agents
Unify conversations with CRM, ticket, product-usage, billing, and other business events that change over time.
Agents working with evolving facts
Track both current and superseded relationships when a flat chat buffer or static document-retrieval system loses important temporal context.
Engineering teams that want a managed memory layer
Use hosted ingestion, graph construction, retrieval, observability, and production deployment controls without assembling the entire graph stack internally.
Capabilities
Core Zep features
Temporal Context Graph
Extracts entities, relationships, facts, and source episodes while recording when facts became valid or were invalidated.
High-level Memory API
Stores session messages and returns a context string, relevant facts, and recent messages for the next model call.
Low-level Graph API
Adds text or structured business data and searches nodes or edges when teams need more control than the opinionated Memory API provides.
User and shared graphs
Maintains a graph across all sessions for an individual user or creates group graphs for shared organizational and domain knowledge.
Context assembly
Builds token-conscious context blocks from relevant summaries, facts, entities, episodes, and other context types, with customizable templates on supported plans.
Custom graph structure
Supports custom entity and edge types plus extraction instructions so graph construction can reflect a product's domain.
Deployment and governance options
Enterprise options include managed cloud, customer-managed encryption keys, bring-your-own-cloud deployment, audit controls, retention policies, and data-access controls.
Graphiti open-source framework
Provides a separately operated Python framework and MCP server for teams that want to build temporal knowledge graphs on supported graph databases.
Process
How the Zep workflow works
Step 1
Define the memory boundary
Decide whether each graph represents a user, account, team, or topic, and specify which applications and roles may read or write it.
Step 2
Create users and sessions
Provision identifiers and add enough authorized identity metadata for messages and business data to map to the correct subject.
Step 3
Ingest conversations and business data
Send ordered chat turns plus relevant text or JSON episodes from approved sources, including timestamps and provenance where available.
Step 4
Retrieve context safely
Use the Memory API for a prepared context block or the Graph API for custom search, and pass retrieved content to the model as untrusted data rather than privileged instructions.
Step 5
Evaluate and govern production use
Test fact extraction, invalidation, retrieval quality, latency, isolation, deletion, prompt-injection resistance, and cost with representative workloads before expanding access.
Cost
Zep pricing and free plan
Zep uses ingestion credits based on Episode size. An Episode up to 350 bytes costs one credit, with another credit for each additional 350 bytes or part; retrieval, storage, graph storage, memories, and users are unmetered. Prices below are the monthly self-serve rates shown on September 1, 2026. Annual billing advertises a 17% saving, and enterprise pricing is negotiated.
Free
$0 for prototyping
A limited monthly allowance for testing Zep before choosing a paid plan.
- 10,000 credits per month
- 2 projects and 1 Memory MCP Server seat
- 5 custom entity and edge types
- Variable rate limits and lower-priority Episode processing
- No credit rollover or automatic top-up
Flex
$125/month
The entry self-serve production plan with automatic credit top-ups.
- 50,000 credits per month
- $25 per additional 10,000 credits
- 600 requests per minute, 5 projects, and 5 Memory MCP Server seats
- 10 custom entity and edge types
- 30-day credit rollover and 1 day of API logs
Flex Plus
$375/month
Higher-volume self-serve access with more graph customization and operational features.
- 200,000 credits per month
- $75 per additional 40,000 credits
- 1,000 requests per minute, 10 projects, and 15 Memory MCP Server seats
- Observations, custom extraction instructions, webhooks, and analytics
- 60-day credit rollover and 7 days of API logs
Enterprise
Custom
Negotiated capacity, security, support, and deployment for production workloads.
- Custom credits and guaranteed rate limits with an SLA
- Unlimited projects and custom Memory MCP Server seats
- SOC 2 Type II materials, HIPAA BAA availability, and one year of audit and API logs
- Managed cloud, bring-your-own-key, or bring-your-own-cloud deployment
- Dedicated account support
Pricing checked . Check current pricing at the source ↗
Assessment
Zep strengths and limitations
Where it stands out
- Models changing facts explicitly instead of treating memory as an undated collection of similar text
- Combines conversations and structured business events in the same user or group graph
- Offers both a simple prepared-context API and lower-level graph search for custom retrieval
- Preserves source episodes and provenance so teams can trace extracted context back to inputs
- Provides supported SDKs for Python, TypeScript, and Go without locking applications to one agent framework
- Separates a managed production service from an Apache-2.0 open-source framework for teams with different operating preferences
- Enterprise deployment, encryption-key, compliance, retention, audit, and access-control options address more demanding data boundaries
What to consider
- Zep is developer infrastructure, not a finished general-purpose chatbot; teams must build the agent, prompts, model calls, tools, identity controls, and product experience around it
- AI-extracted entities and facts can be incomplete, incorrectly merged, or wrongly invalidated, so domain-specific evaluation and human review are still necessary
- The official Memory API documentation says ingestion can take a few minutes; applications should include recent raw messages so the latest turns are not lost
- Usage cost depends on the byte size and number of ingested Episodes, while downstream model, embedding, infrastructure, and application costs may also apply
- Retrieved chat, document, and tool content is untrusted input and can contain prompt-injection attempts; it should never be inserted as a privileged system or developer instruction
- Sensitive data requires deliberate minimization, tenant isolation, retention, deletion, consent, and access-control design; a compliance certification does not make an application compliant by itself
- Several governance and compliance capabilities are enterprise features, including HIPAA BAAs, expanded logs, negotiated SLAs, and private deployment options
- Graphiti self-hosting shifts graph-database operation, LLM and embedding configuration, scaling, security, backups, monitoring, and upgrades to the adopter
Compare
Zep alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
LangChain
Choose LangChain when you want a broad open-source framework for composing agents and prefer to assemble memory, retrieval, storage, and evaluation components yourself.
Explore LangChain →Data Analysis
Pinecone
Choose Pinecone when the core requirement is managed vector search and conventional RAG rather than an automatically constructed temporal knowledge graph.
Explore Pinecone →Business Operations
n8n
Choose n8n when the primary problem is connecting applications and orchestrating AI workflows; it can coordinate external storage but does not replace Zep's specialized temporal memory layer.
Explore n8n →Questions
Zep FAQs
What does Zep do?
Zep ingests conversations and business data, constructs temporal context graphs, and returns relevant memory or graph-search results that an application can provide to an AI agent.
Is Zep a vector database?
Not primarily. Zep combines semantic and full-text retrieval with graph relationships, time-aware facts, source episodes, and context assembly. A vector database is a simpler alternative when document similarity search is the main need.
Is Zep open source?
The current hosted Zep platform is a commercial service. Graphiti, the temporal knowledge-graph framework associated with Zep, is open source under Apache 2.0 and must be deployed and operated separately.
How much does Zep cost?
As reviewed September 1, 2026, prototyping includes 10,000 free credits per month. Flex is $125 per month with 50,000 credits, Flex Plus is $375 per month with 200,000 credits, and Enterprise is custom-priced. Ingestion credits scale with Episode size.
Which programming languages does Zep support?
Zep documents official SDK quickstarts for Python, TypeScript, and Go. Its API can work with an agent framework or a custom application.
Does Zep replace recent chat history?
No. Zep's documentation recommends sending recent messages alongside the retrieved long-term context because newly ingested information may not yet appear in the graph.
Can Zep handle sensitive or healthcare data?
Zep advertises SOC 2 Type II certification and offers HIPAA Business Associate Agreements on Enterprise, plus BYOK and BYOC options. A team must still perform its own legal, security, privacy, data-flow, access, and retention review before sending sensitive data.
Can Zep memory be wrong?
Yes. Entity extraction, relationship resolution, temporal invalidation, and retrieval are model-assisted processes. Test representative cases, preserve provenance, add access checks, and require human review wherever an incorrect fact could cause material harm.
Bottom line
Our Zep verdict
Zep is a strong fit when an AI product needs durable, evolving context across conversations and business systems and the team wants a managed memory layer rather than a hand-built retrieval stack. Its temporal graph, source provenance, high- and low-level APIs, and enterprise deployment controls are meaningful advantages. It is more infrastructure—and potentially more cost and governance work—than a simple chatbot or vector-search project needs, so validate extraction quality, latency, prompt-injection defenses, deletion behavior, and credit consumption with a production-shaped pilot.
Visit Zep website ↗Similar Coding Tools
View all Coding AI tools →
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.