Zep

About Zep

Zep is a managed memory layer for AI agents built on temporal context graphs rather than a vector store with a similarity search bolted on. The distinction matters for the failure it addresses: an agent that retrieves the most semantically similar past message will happily surface a fact that was true six months ago and has since been superseded. Zep's graph tracks facts across time, so a superseded fact is recorded as superseded rather than silently competing with the current one, and it ingests chat history, business data and user behaviour into the same structure. The open-source foundation, Graphiti, crossed 20,000 GitHub stars in under twelve months, which gives the approach a public implementation to inspect rather than only a marketing claim. Around the graph sit the Context Lake and the Context Graph Engine, with sub-200ms retrieval quoted as the latency budget — the number that decides whether a memory layer can sit in the request path of a conversational agent at all. The company publishes benchmark results on LoCoMo and LongMemEval, the two standard long-running-agent memory benchmarks, rather than asserting quality. Compliance is SOC 2 Type II with a HIPAA BAA available, and deployment can be Zep's cloud or your own VPC. Self-serve plans are credit-metered with 30-day rollover and auto top-up, and the company offers enterprise terms at emerging-company pricing for funded startups.

Key Features

Tags