Zep

What does Zep do?

Zep assembles optimized context from chat history, business data, and user behavior so agents can respond with the right information at the right time. Its context assembly flow pulls from memory and business records, then formats the result into reusable context blocks with Dynamic Relevance Ranking, Memory Context Blocks, and Context Templates. For teams that need more control, the Graph Search API and Custom Entity and Edge Filtering let you shape exactly what gets surfaced without hand-crafting prompts. At scale, Zep is built for low-latency retrieval and changing data: the site shows under-200ms retrieval, <200ms P95 retrieval latency, and real-time incremental updates instead of batch recomputation. It supports a unified context graph for evolving facts, and the open-source Graphiti layer reports 100%+ accuracy improvements, 90% latency reduction, and 98% fewer tokens required for processing. Customers shown on the site include Twin Health, Praktika.ai, Writer, Samsung, and HoneyBook.

Why use Zep?

• It replaces manual prompt crafting with automated context assembly, reducing the work needed to keep agent inputs relevant.
• Its temporal graph approach keeps historical context and current facts together, which helps agents handle changing user and business states.
• Low-latency retrieval under 200ms supports voice and other latency-sensitive experiences.
• Enterprise deployment options include managed, BYOK, BYOM, and BYOC, giving teams control over security and infrastructure.
• The platform supports real-time incremental updates, so changing records do not require batch recomputation.

Who is Zep for?

• AI product teams who need agents to remember users across sessions and interactions.
• Developers building context-aware workflows who want graph-based retrieval instead of manual prompt assembly.
• Engineering leaders who need low-latency, production-ready context infrastructure for dynamic business data.
• Teams handling sensitive data who need flexible enterprise deployment and compliance controls.

What are Zep's key features?

What does Zep integrate with?

• LangChain
• LangGraph
• OpenAI
• Anthropic
• AWS KMS
• CloudTrail
• Google
• Azure
• Claude
• Cursor
• Neo4j
• Azure OpenAI
• Google Gemini

What are Zep's use cases?

How does Zep work?

1. Connect your first data source or chat stream, then define what Zep should remember using Agent Memory and Context Templates. Start with the conversations, records, or events that matter most.
2. Map entities and relationships with Custom Entity Types and Custom Entity and Edge Filtering, so Zep can build a useful graph instead of storing raw text alone.
3. Let Intelligent Context Assembly and Dynamic Relevance Ranking select the right facts for each request, reducing manual prompt assembly and keeping token usage focused.
4. Use the Graph Search API or Framework Integration with LangChain, LangGraph, or OpenAI to wire context into your app's runtime and agent workflows.
5. Monitor retrieval quality, latency, and updates in the API logs and analytics, then refine templates, filters, and data sources as your product evolves.

How much does Zep cost?

Frequently asked questions