Zep
A memory layer for AI assistants and agents. Zep provides long-term memory, conversation history management, and user context for LLM applications. It automatically extracts facts, summarizes conversations, and enables semantic search over past interactions.
Implements
Concepts this tool claims to implement:
- Long-Term Memory primary
Persistent memory storage with automatic fact extraction. Entity recognition and relationship tracking. Memory retrieval based on relevance and recency.
- Agent Memory primary
Conversation history with summarization. Session management for tracking user interactions. Memory injection into prompts.
- Semantic Memory primary
Semantic search over conversation history. Vector embeddings for memory retrieval. Fact and entity extraction from conversations.
- Episodic Memory secondary
Conversation sessions as episodes. Temporal ordering of interactions. Session-based memory organization.
Integration Surfaces
Details
- Vendor
- Getzep Inc.
- License
- Apache-2.0 (OSS) / Proprietary (Cloud)
- Runs On
- cloud, local
- Used By
- system, agent
Links
Notes
Zep focuses on being a drop-in memory layer rather than a full agent framework. Integrates with existing stacks (LangChain, LlamaIndex). Open-source version can be self-hosted, cloud version adds more advanced features. Good for adding memory to existing applications.