Zep

Tier 2: Persistent (session-scoped)

Temporal knowledge graph for agents that need to know when

getzep.com ↗ · ~14,000 (Graphiti) GitHub stars · Undisclosed

Our take

Zep is the most interesting tool in this space, at least technically. The temporal knowledge graph is a novel idea. No other tool natively tracks when facts were true, when they were learned, and how they changed. The Graphiti open-source library is well-designed. But here's the reality: the SaaS offering is still rough. Their infrastructure broke under a 30x usage spike. If you want to learn where AI memory is heading, Zep is worth studying. If you need something in production next month, look elsewhere first.

How it works

Temporal knowledge graph (Graphiti engine). Every edge tracks when an event occurred and when it was ingested, with explicit validity intervals. Combines semantic embeddings, keyword search, and graph traversal.

When to use Zep

  • + Applications where "when did we learn this" matters as much as "what do we know" (compliance, audit trails, legal tech)
  • + Agents that need to reason about how facts have changed over time
  • + Teams that want an open-source graph engine they can self-host and customize

When to skip it

  • − Simple key-value memory (massive overkill)
  • − Teams without Neo4j experience (the dependency is non-trivial)
  • − Anyone expecting a polished SaaS experience right now

What it does better than everything else

Temporal reasoning. No other tool natively answers "what did we know about X as of last Tuesday?", tracking how facts and relationships evolve over time.

MCP support

Supported. Graphiti MCP server available. Provides episode management, entity management, semantic search, and temporal queries through MCP. Requires Neo4j and Docker.

Why MCP matters: 5 MCP servers every engineering team should run →

Pricing

Free

API access with lowest rate limits, no credit card

Paid

Credit-based: 1 Episode = 1 credit. Storage is free. Auto-top-up when balance drops below 20%.

The gotcha nobody mentions

You must run and manage a Neo4j instance. If you've never operated a graph database in production, expect a steep learning curve on top of learning Graphiti itself. Docker Compose setup failures are common in the GitHub issues.

"Look into Zep if your primary goal is to learn about the cutting edge of AI memory systems. It's a valuable resource for understanding advanced concepts, even if their product isn't ready for prime time deployment yet."

Frequently asked

Is Zep the same as Graphiti? +
Graphiti is Zep's open-source temporal knowledge graph engine. Zep Cloud is the managed SaaS built on top of it. You can use Graphiti standalone without Zep Cloud.
Can I use Zep without Neo4j? +
Technically Graphiti supports FalkorDB and Neptune, but the codebase is tightly coupled to Neo4j internals. In practice, Neo4j is the only reliable option.
How does Zep compare to Mem0 for simple chatbot memory? +
Mem0 is simpler and faster for basic chatbot memory. Zep is overkill unless you specifically need temporal reasoning, knowing how facts changed over time.

Related reading

→ The 3 types of AI memory → 5 MCP servers for teams

Also in this space

Mem0

The fastest path from zero to deployed memory

Cognee

Knowledge graphs for AI that needs to connect the dots

Letta

Agents that manage their own memory (formerly MemGPT)

Zep is Tier 2 — persistent (session-scoped) memory.

It solves real problems for individual agents and users. But if your team needs shared, always-current memory that works across Cursor, Claude, and every other AI tool simultaneously — that's a different architecture entirely.

We're building that with Knowledge Plane. Join the beta →