Side by side
Competitor claims last verified 1 November 2026| Dimension | NOAN | Zep |
|---|---|---|
| Purpose | Holds your company's own approved facts, like prices, policies and product details, so every agent reads the current version | A context layer that builds temporal knowledge graphs from business data, documents and conversations |
| Source of truth | One current fact per block, written and verified by your team. A new version supersedes the last | Facts an LLM pipeline extracts from what you add. When new data contradicts a fact, the older fact is invalidated |
| Who verifies | A named person on your team. Facts an agent writes can wait for a person to verify them before anything reads them | Its docs describe no human approval step for facts. Enterprise content policies (beta) can repair or drop extracted items automatically |
| Versioning and history | Every version kept, with who verified it and when. Time Travel shows your facts as they stood on any past date | Each fact carries when it became valid and when it stopped, inferred by Zep, so you can ask what was true at a past time |
| Permissions | Roles and fact permissions decide who sees and edits each area. Each API key is scoped further: read or write per resource, a list of allowed stacks, and fact changes held for approval | User and shared graphs within a project. Role-based dashboard access, default-deny access for API keys and user groups, audit logs and SSO |
| How agents access it | Remote MCP server (OAuth or an API key), REST API with an OpenAPI spec, and an open Agent Skill. Plain JSON from any language, no SDK needed | REST API, Python, TypeScript and Go SDKs, a hosted Context MCP server, and integrations such as LangGraph, CrewAI and the Vercel AI SDK. Its graph engine, Graphiti, is open source |
| Where it fits | The approved facts about your business that the rest of the context defers to | Assembling the right context about each user, account or project from large, changing data |
When to choose Zep
- You have large volumes of conversations, documents and business data, and agents need the relevant slice of it
- You want a graph per user, customer or account that updates itself as data arrives
- You need to ask what was true about a customer at a past point, worked out from the data
- You want the open-source Graphiti engine, or SDKs in Python, TypeScript and Go
When to choose NOAN
- A fact has to be the one your team approved, not the one inferred from the latest message
- Agents quote your prices, policies or product details to customers
- You need a named person behind every fact, and every version kept
- Agents, apps and your website should all read the same facts over MCP or REST
The deeper comparison
Inferred validity and verified facts
Zep is careful about time. Each fact in its graph carries when it became valid and when it stopped, which Zep infers from the data, and its docs say contradictions invalidate older facts rather than leaving conflicting truths side by side. When the truth about a customer lives in a stream of calls, emails and records, that is a strong design.
Inference works from whatever data arrives, so the newest evidence decides what is current. That is the right behavior for context about a customer. For your own prices and policies, you want the current version to change only when your team says so. NOAN starts from that end: a person verifies the fact, and agents read the verified version until someone verifies a new one.
Using Zep and NOAN together
Zep for context about users, accounts and projects. NOAN for the approved facts about your business. The agent asks Zep what it should know about this customer, and asks NOAN what your company says is true.
Add one rule to the system prompt: when the two disagree about the business, NOAN wins. Both offer hosted MCP servers, so the same agent can connect to both.
Common questions
Is NOAN an alternative to Zep?
Not for what Zep does best. Zep assembles context about users and accounts from large, changing data. NOAN holds your company's own approved facts with a named verifier and full version history. Teams that need both run both.
Can Zep be the source of truth for my company's facts?
Zep's docs describe facts extracted by an LLM pipeline, with validity dates Zep infers and no human approval step. That suits context. For facts your team must approve, like prices and policies, a fact layer such as NOAN is the better fit, and Zep can carry everything else.
What's the difference between context engineering and a fact layer?
Context engineering is assembling the right information for an agent at the right moment, which is what Zep is built for. A fact layer is one source that context should draw on: your company's own verified facts, served the same way to every agent.
Do Zep and NOAN work together?
Yes. Use Zep for memory and context about each user or account, and NOAN for the approved facts about your business. Both connect over MCP, and a one-line rule tells the agent that NOAN wins on business facts.
Sources: Zep overview · Zep: context graphs · Zep: facts · Zep: governance · Zep: Context MCP server · Graphiti on GitHub · This page as Markdown · All comparisons
Next step
Connect your agent to verified facts
A remote MCP server for your company's own facts. Sign in over OAuth or send a NOAN API key. Install steps for Claude, ChatGPT, Cursor and Codex.
Plain REST with a bearer key, JSON in and out, from any language. Every endpoint is in the OpenAPI spec.
Or tell any agent: Read https://www.getnoan.com/setup.md and follow it.
See the fact layer with your facts in it
Every plan includes the API and MCP, and paid plans add unlimited keys. Start with one verified fact and build from there.