Graphiti is the open-source Context Graph framework — the engine that turns your data into a temporal Context Graph. Zep is agent memory at enterprise scale: it runs Graphiti inside a managed system — ingestion, retrieval, governance, and the proprietary Context Graph Engine runtime — and serves millions of governed Context Graphs as one Context Lake.
In short: Graphiti builds the graph; Zep operates it at scale.
Graphiti: the Context Graph framework
Graphiti turns conversations, business data, and documents into a temporal Context Graph — entity extraction, relationships, your ontology, and fact invalidation. It builds one Context Graph per subject (a user, customer, team, or topic), runs locally, and provides the data model and a hybrid retrieval API.
- Temporal by design — every edge carries valid-from, valid-to, learned-at, and invalidated-at timestamps. Point-in-time queries follow directly from the data model.
- Hybrid retrieval — vector similarity, full-text search, and graph traversal combine into one ranked answer. No LLM-in-the-loop reranking and no orchestration layer to maintain.
- Dynamic updates — new facts integrate immediately. Outdated facts are invalidated by temporal logic and preserved as history rather than deleted.
- Pluggable backends — graph storage with Neo4j, FalkorDB, or Amazon Neptune; LLM providers including OpenAI, Azure OpenAI, Google Gemini, and Anthropic.
- MCP server — connect Graphiti to Claude, Cursor, and other MCP clients.
How Graphiti fits into Zep
In Zep, Graphiti is the graph-construction layer in a larger pipeline. Zep invokes it through a single SDK and operates the rest of the system around it:
- Agent runtime — any framework (LangChain, LlamaIndex, CrewAI, Google ADK) or none, through one SDK.
- Ingestion — chat, JSON, documents, and app events arrive from any source the agent touches.
- Graphiti — turns that data into a temporal Context Graph: entity extraction, relationships, your ontology, and fact invalidation.
- Context Graph Engine — Zep’s proprietary runtime serves the temporal Context Graph — entities, facts and edges, episodes, and decision traces — at sub-200ms regardless of scale.
- Retrieval and Context Assembly — retrieval is auto-optimized, provenance-linked, and policy-filtered; Smart Context Assembly packs the result into a token-efficient Context Block.
- Governance — ABAC, multi-tenant isolation, customer-key encryption, retention, audit, and provenance apply to every read and write, in the substrate rather than bolted on.
Run Graphiti on its own when you want the framework and will operate the surrounding system yourself. Use Zep when you want that pipeline managed and served at enterprise scale.
Zep: agent memory at enterprise scale
Zep serves millions of governed Context Graphs as one system — a Context Lake — with sub-200ms retrieval regardless of graph size or count. Beyond the pipeline above, Zep adds the managed-service and enterprise layer:
- Built-in storage — users, threads, and message storage are managed for you.
- SDKs and tooling — official SDKs for Python, TypeScript, and Go, plus a dashboard with graph visualization, debug logs, and API logs.
- Compliance — SOC 2 Type II and HIPAA.
- Deployment models — Cloud (managed), BYOK, and BYOC (your VPC).
Comparison
When to choose which
Choose Graphiti if you want a flexible OSS core and you are comfortable building and operating the surrounding system.
Choose Zep if you want turnkey agent memory at enterprise scale — a managed Context Lake with governance, performance, and support baked in.
