What is Zep?
Zep gives AI agents enterprise-grade memory that actually understands how people, businesses, and facts change over time. Unlike basic chat logs or static document retrieval, Zep builds temporal context graphs—automatically connecting conversations, user actions, and business data into a living knowledge base that stays accurate and up to date.
Built for real-world enterprise needs, Zep delivers sub-200ms retrieval even at massive scale, with full governance, audit trails, and SOC 2 compliance. Trusted by teams at companies like Flockx and Axtria—and validated by S&P Global Market Intelligence—it’s designed so your AI agents remember what matters, forget what’s outdated, and always act on the latest truth.
What are the features of Zep?
- Temporal Context Graphs: Automatically track how facts evolve over time (e.g., “Robbie loved Adidas” → “Robbie switched to Nike”) and keep historical context intact.
- Sub-200ms Retrieval at Scale: Consistently fast memory access—even with millions of context graphs—thanks to Zep’s proprietary Context Graph Engine.
- Context Lake Architecture: Ingest and unify memories from any source—chats, CRM data, support tickets, events—into one governed system, like a data lake for agent memory.
- Automatic Observations: Go beyond raw facts; Zep detects patterns (e.g., “Jane upgrades within 2 weeks of every product launch”) to give agents deeper insight.
- Enterprise Governance: Built-in access control, retention policies, legal hold, and full audit logs—applied at the substrate level, not as an afterthought.
- Provenance Tracking: Every fact traces back to its original source (e.g., a chat message or order return), so you can audit any AI response.
- Multi-Cloud Deployment Options: Choose Zep Cloud (fully managed), BYOK (bring your own encryption keys), or BYOC (deploy in your VPC) based on your security needs.
What are the use cases of Zep?
- Power personalized customer support agents that remember past interactions, preferences, and issue history across channels.
- Build investment or portfolio review assistants that track evolving client goals, risk profiles, and market commentary over time.
- Enable sales agents that recognize buying patterns (e.g., frequent upgrades) and trigger timely outreach.
- Create internal ops agents that combine HR data, project updates, and meeting notes to answer employee questions accurately.
- Develop compliance-aware agents that respect data retention rules and access policies while retrieving context.
- Support long-running research agents that synthesize insights from months of documents, calls, and observations.
How to use Zep?
- Add user messages and get relevant memory in one call using
add_messages(..., return_context=True). - Inject structured business data (e.g., plan upgrades, orders) directly into a user’s context graph via the
graph.add()API. - Retrieve concise, token-efficient context for your agent using
get_user_context(thread_id). - Use the Batch API to load large historical datasets (up to 50k items per batch) without disrupting real-time performance.
- Monitor memory activity, latency, and error rates through Zep’s built-in observability dashboard.
- Apply retention policies or legal holds through the Trust Center to meet compliance requirements automatically.
