Zep AI: Agent Memory and Enterprise Context, Explained
Zep gives AI agents memory and context built on knowledge graphs. It started as a memory layer that remembers each user across conversations, and now calls itself “the enterprise context layer for reliable agents”: it connects conversations, documents and business data into governed graphs that agents can query in under 200 milliseconds. Its distinguishing feature is time. When a fact changes, Zep keeps the old one with an end date, so an agent knows both what’s true now and what was true before. Zep is built on Graphiti, its open-source graph framework, which has about 32,800 stars on GitHub.
The quick answer: Zep is one of the most enterprise-focused AI memory companies, with access policies, audit logs, SOC 2 Type II, HIPAA and deployment in your own cloud. Its temporal graphs suit agents that need to reason about facts that change, like customer accounts or company policies. A disclosure: we make Carly, an AI assistant with its own memory and workspace, so we compete on the memory side. Everything below comes from Zep’s site, blog, docs and Y Combinator profile.
Zep at a glance
Checked against Zep’s site, blog, pricing and Y Combinator profile on November 12, 2026.
| Zep | |
|---|---|
| What it is | A context and memory layer for AI agents, built on temporal knowledge graphs |
| Open source | Graphiti, Apache 2.0, about 32,800 GitHub stars |
| Speed | Sub-200ms retrieval, on Zep’s own graph database service, Konig |
| Connectors (Enterprise) | Google Drive, Slack, Notion, Gmail, Granola, Microsoft Teams, SharePoint, Snowflake, Databricks, BigQuery |
| Deployment | Zep Cloud, Zep Cloud with your encryption keys, or your own VPC |
| Compliance | SOC 2 Type II and HIPAA BAA |
| Company | Founded 2023 by Daniel Chalef; Y Combinator W24 |
What Zep does
Zep’s homepage describes “governed access to company knowledge and skills built from agent tasks.” In practice it offers three things:
- Unified business context. Through what it calls the Context Lake, agents get “a complete and current understanding of your business at runtime,” with facts linked to their sources.
- Self-improving agents. “Agents get more tasks right and use fewer tokens with skills Zep builds from their runs.” This newer feature, Agent Skills, is free on every plan until January 4, 2027.
- Governance as infrastructure. “Policies decide what each user and agent can read, with every request logged.”
Its listed uses include a company brain for internal agents, a unified view of each customer for product agents, memory of each user that stays current, and agents for regulated industries like legal, finance and healthcare, where each customer gets its own isolated graph.
How the graphs handle time
Zep’s core technology is the temporal context graph, described in its research paper “Zep: A Temporal Knowledge Graph Architecture for Agent Memory.” Instead of storing text chunks, it stores entities and the relationships between them.
The time dimension is what sets it apart. According to a recent Zep post, each fact has two timelines: “The event timeline records when the fact was true, and the system timeline records when the graph learned the fact and when it learned that the fact had ended.” When a new fact contradicts an old one, Zep doesn’t delete the old fact. It gives it an end date, so an agent can still answer questions about the past.
Zep’s example: a data migration in September 2025 changes the key used to join orders to accounts. An analyst later asks for revenue for the first quarter of 2025. A system that trusts the newest information would use the new key and get the wrong answer. Zep knows the old key applied until September, and uses it.
The newest version of its API, Zep v4, added features such as custom entity type hierarchies and “hyperedges” for facts that involve three or more entities at once.
Memory across the AI tools people already use
In August 2026, Zep launched a Memory MCP server that opens the same memory to Claude, Claude Cowork, ChatGPT, Cursor and other MCP clients. Zep’s example: “A preference recorded in Claude is applied by an agent you built, then recalled in ChatGPT.”
Because Zep is aimed at companies, the MCP server is tied to identity. Each person signs in through Google Workspace or the company’s identity provider and reaches only their own memory, while company-wide graphs are available according to group policies. Zep also offers a plugin that helps Claude Code, Codex and Cursor design Zep implementations.
Open source: Graphiti, not Zep
Zep’s open-source story changed in 2025. The company used to offer an open-source Zep Community Edition alongside its cloud service. In April 2025, founder Daniel Chalef announced it would stop maintaining that edition, saying the company “felt uneasy about the common practice in open-core of intentionally limiting features to drive users toward paid products.” The old repository stays available but unsupported.
Zep’s open-source work now centers on Graphiti, the temporal knowledge graph framework that powers the service, licensed Apache 2.0. It’s one of the most-starred projects in agent memory, with about 32,800 stars. Teams that want to self-host build on Graphiti directly; teams that want the managed service, governance and connectors use Zep.
Pricing
Zep’s pricing is credit-based:
| Plan | Price | What it includes |
|---|---|---|
| Free | $0 | 10,000 credits a month, with limits |
| Flex | $125 a month | 50,000 credits, unlimited memories, retrieval and users |
| Flex Plus | $375 a month | 200,000 credits, custom extraction instructions, webhooks, analytics |
| Enterprise | Custom | Your encryption keys or VPC, SLA, SOC 2 Type II and HIPAA BAA, connectors, year-long audit logs |
Zep also offers enterprise plans at reduced prices for fast-growing venture-backed startups.
The company
Zep was founded in 2023 by Daniel Chalef, its CEO, and went through Y Combinator’s Winter 2024 batch. In April 2026, S&P Global Market Intelligence’s 451 Research published analyst coverage titled “Zep tackles agent memory limitations through its temporal context graph,” which Zep quotes as saying “We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack.”
How Zep compares
Zep competes with other memory layers, including Supermemory, Honcho, Mem0 and Letta. Its differences are the temporal graph that tracks when facts were true, a stronger focus on company data and governance than on personal memory, and a large open-source framework in Graphiti. Supermemory leans toward broad content ingestion and Honcho toward reasoning about individual users. Our guide to connecting Mem0 to an AI agent covers another rival.
Zep is built for companies wiring memory into their own agents. If you want an assistant that already remembers you, Carly has memory built in: it saves named memory documents that it reads and searches in later conversations, searches a knowledge base of files you upload by meaning, and keeps your to-dos, data tables and CRM in one workspace. That memory comes with you whether you use Carly inside ChatGPT, inside Claude or by text, and Carly also labels and sorts email across all your accounts, handles advanced scheduling, and makes and answers calls and texts.
Who Zep suits
Zep is a good fit if:
- You’re building agents that need current, governed context about customers or company data.
- Your facts change over time and agents need to know what was true when.
- You need SOC 2 Type II, HIPAA, audit logs or deployment in your own cloud.
- You want one memory shared between your own agents and tools like Claude and ChatGPT.
Look elsewhere if:
- You want a fully open-source memory server; Zep’s open source is the Graphiti framework, not the service.
- You only need lightweight memory for a hobby project.
- You want a finished assistant rather than infrastructure to build one.
FAQ
What is Zep?
Zep is a context and memory layer for AI agents. It builds temporal knowledge graphs from conversations, documents and business data, and gives agents governed, up-to-date context in under 200 milliseconds.
What is Graphiti?
Graphiti is Zep’s open-source temporal knowledge graph framework, licensed Apache 2.0, with about 32,800 GitHub stars. It powers the Zep service and can be used on its own.
Is Zep open source?
Zep’s service isn’t open source. Zep stopped maintaining its open-source Community Edition in April 2025 and now focuses its open-source work on Graphiti.
How much does Zep cost?
Zep has a free tier with 10,000 credits a month, Flex at $125 a month, Flex Plus at $375 a month and custom Enterprise plans.
Does Zep work with Claude and ChatGPT?
Yes. Zep’s Memory MCP server connects Claude, Claude Cowork, ChatGPT, Cursor and other MCP clients to the same user memory as agents built on Zep.
Who founded Zep?
Daniel Chalef founded Zep in 2023. It’s a Y Combinator Winter 2024 company.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
