Zep

What is Zep?

Zep is an open-source platform that provides fast, scalable building blocks for applications built on large language models (LLMs). It lets you move from prototype to production without rewriting code. Features include conversation memory, dialogue classification, data extraction, vector search, metadata filtering, named entity extraction, intent analysis, automatic embeddings, chat history recall, and archival and enrichment capabilities. Zep also supports privacy rules and runs faster than typical offerings from major LLM providers.

Who created Zep?

Zep was built by an open-source community. It focuses on privacy compliance and provides components for memory, search, and enrichment, along with record keeping for compliance and user session management.

What is Zep used for?

• Meet records retention requirements by archiving every user message and AI response.
• Manage users and their chat sessions.
• Work with many languages and frameworks, including Python, TypeScript, LangChain, and LlamaIndex.
• Perform vector searches for semantic information retrieval across chat histories and documents.
• Filter search results with metadata, and use named entity extraction and intent analysis output.
• Generate embeddings automatically with local, low-latency models, or bring your own vectors (BYOV).
• Fill prompts with relevant chat history and automatically enrich messages for building agent tools.
• Comply with corporate and regulatory record-keeping mandates, including privacy rules like CCPA and GDPR.

Who is Zep for?

Zep is aimed at developers, software engineers, data scientists, chatbot developers, AI researchers, linguists, and compliance officers.

Key capabilities

• Personalized experiences: Build low-latency agents that retrieve relevant facts from past conversations.
• Structured data extraction: Pull structured data from chat messages for business processes, user profiles, and API calls.
• Compliance and governance: Archive conversations and manage user interactions to satisfy retention and privacy requirements.