ZepZep Python is the Python client for Zep, a context-engineering platform for AI assistant applications. Zep persists chat histories and recalls them however far back they run, generating summaries and other artifacts asynchronously so the work does not sit in the user's chat path, and embedding messages and summaries so relevant past context can be searched.
With it you assemble a prompt deliberately rather than replaying a whole history: a set number of recent messages, a summary of the conversation before them, contextually relevant summaries or messages drawn from the entire session, and business data held in Document Collections — a vector-search abstraction meant to complement the chat features rather than serve as a general-purpose vector database.
Three installation targets cover the different deployments: a zep-cloud package for Zep Cloud, zep-python for the Community Edition, and a pinned 1.x range for the v0.x-compatible client. A Zep project API key authenticates the client. Cloud adds fact extraction into tables without a predefined schema, dialog classification for intent and emotion, and structured data extraction against a schema you define.
Pre-release versions ship ZepChatMessageHistory and ZepVectorStore classes importable from zepcloud.langchain for use with LangChain's expression language, which require langchaincore in the 0.1.x range.
Features
- Chat-history recall: past messages are persisted and retrievable regardless of how far back they occurred
- Automatic summarisation: summaries and other artifacts are generated from chat histories asynchronously
- Semantic search: messages and summaries are embedded so relevant context can be surfaced from a whole session
- Prompt construction: recent messages, prior-conversation summaries, and retrieved context are combined per request
- Document Collections: a vector-search abstraction for business data alongside the chat-history features
- Fact extraction: Cloud builds fact tables from conversations without a schema defined upfront
- Dialog classification: Cloud classifies chat dialog for intent, emotion, user segmentation, and routing
- Structured data extraction: business data is pulled from conversations against a schema you supply
- LangChain classes: ZepChatMessageHistory and ZepVectorStore for expression-language pipelines
