Graphiti vs Memlo: which to use for agent memory

Last updated October 31, 2026

Graphiti is an open-source Python framework, from Zep, for building temporal knowledge graphs: it turns text into entities and facts, tracks when each fact was true, and runs on a graph database you host, such as Neo4j or FalkorDB. Memlo is a hosted memory server for the agents you already use: Claude Code, Codex, claude.ai, ChatGPT and other MCP apps share memories that agents save on purpose, each with its own version history. Pick Graphiti to build graph memory into your own agent or app and host it yourself. Pick Memlo to give your own agents and your team one shared memory with nothing to run.

At a glance

Facts about Graphiti come from its own repository and documentation, read on October 31, 2026. Links are in Sources at the end.

How Graphiti works

You add "episodes" (messages, documents or JSON) to Graphiti. A language model reads each one and pulls out entities and the relationships between them, and Graphiti stores them in a graph with their time ranges. In its own words, it offers "explicit bi-temporal tracking with automatic fact invalidation": when a fact changes, "old facts are invalidated — not deleted," so you can ask what is true now or what was true at an earlier time. Search combines the graph with text and vector search.

Graphiti needs a graph database (Neo4j, FalkorDB or Amazon Neptune; Kuzu support is deprecated) and a model provider. It defaults to OpenAI for both extraction and embeddings, and its README says it depends on models that reliably return structured (JSON) output.

Graphiti has an MCP server, but its README calls it a "reference implementation" and says not to use it in production: "The server has no authentication, no authorization, and no prompt-injection protection." It runs on your computer with FalkorDB or Neo4j, over HTTP or stdio.

How Memlo works

Memlo is an MCP server at https://memlo.ai/mcp. You add it to each agent and sign in once, with no API key. Agents get tools to save, search, read, change and retire memories.

• Curated memories. Each memory is a short note with a name and a title, saved when the agent learns something worth keeping.
• History, with authors. Every change is a version, with when it was written and by whom. Any version can be restored, and edit_memory refuses a write when someone else changed the note first.
• Spaces and sharing. Keep memories in spaces, and share a space or a page with people by email as a Viewer or an Editor.
• A wiki and a graph. Memlo keeps a page for each person, project and tool your memories mention, with every line citing its memory.

Where Graphiti is the better choice

• You're building an agent or app. Graphiti is a library you call from your own Python code. Memlo has no SDK.
• You need to self-host. Graphiti runs entirely on your own infrastructure, with the database and model you choose. Memlo is hosted only.
• You need questions about time. Graphiti's bi-temporal model answers "what was true on this date" across the whole graph.
• You want to change the code. Graphiti is Apache 2.0.
• You want to start today. It's open source; Memlo is by invitation.

Where Memlo is the better choice

• You want memory for the agents you already use, with no code. Add one URL to Claude Code, Codex, claude.ai or ChatGPT and sign in. There's no database, model key or server to run.
• You want a production MCP server with sign-in. Memlo uses OAuth and checks every call; Graphiti's MCP server is a reference with no authentication.
• You want to read and correct each memory. Memories are short titled notes you can open, edit and roll back, rather than facts extracted by a model.
• You want to see which agent wrote what. Every version names the person and the connected app or agent.
• You share with people. Share a space or a single page by email with a role.

What Memlo doesn't do yet

• Cursor. Memlo's sign-in doesn't work from Cursor yet.
• Automatic capture. Agents save memories on purpose; nothing records sessions in the background.
• An SDK or API keys. Memlo is reached over MCP with your own sign-in, while Graphiti is a library for your own code.
• Self-hosting. Memlo is hosted only.
• Open sign-up. Accounts are by invitation for now; you can join the waitlist.

Frequently asked questions

Is Graphiti open source?

Yes. Graphiti is under the Apache 2.0 license. You run it yourself with your own graph database and model provider.

What is the difference between Graphiti and Zep?

Graphiti is the open-source framework. Zep's documentation calls it "the open-source framework for building temporal knowledge graphs — Context Graphs — that power Zep." Zep is the hosted, managed product. See Zep vs Memlo.

Does Graphiti have an MCP server?

Yes, in its repository. Its README says it is a reference implementation, not for production, with no authentication or authorization. It needs FalkorDB or Neo4j and a model API key.

What is a knowledge graph memory MCP server?

An MCP server that stores memory as entities and the relationships between them. Graphiti's MCP server is one. The MCP project's own reference "Memory" server is a simpler one: "a basic implementation of persistent memory using a local knowledge graph," kept in a file on your computer. Memlo also builds a graph of the people, projects and tools your memories mention, with a page for each.

Is there a Graphiti alternative for Claude Code and Codex?

Memlo is one if you want hosted memory for those agents rather than a framework: Claude Code, Codex and chat apps connect to the same server. See also Cognee vs Memlo and Mem0 vs Memlo.

Sources

Checked on October 31, 2026:

• Graphiti on GitHub and its license
• Graphiti MCP server
• Graphiti documentation
• MCP reference Memory server
• Memlo: the setup page, how it works and terms