Compare

khive vs mem0 vs Zep

khive, mem0 and Zep each give an AI agent a memory that lasts between sessions, and each runs a hosted MCP server that a client reaches at one URL. They differ in what the store holds, in how content gets into it, and in price. The mem0 and Zep facts on this page come from their pricing and documentation pages as they read on 16 November 2026. Vendors change plans often, so read their pages before you decide.

At a glance

Hosted MCP endpoint

khive
https://khive.ai/v1/mcp
mem0
https://mcp.mem0.ai/mcp
Zep
https://api.getzep.com/mcp

Sign-in

khive
OAuth in the browser, or an API key
mem0
Browser authorization, or an API key as a bearer token
Zep
Google Workspace by default; a custom OIDC provider on the Enterprise plan

What the store holds

khive
Memories, entities and notes linked by typed edges, tasks, reference documents
mem0
Memories scoped to a user, agent, app or run. A graph is on the Custom plan
Zep
A temporal graph: the episodes you send, the entities in them, the facts between them, summaries

Task tools

khive
Yes
mem0
None in its MCP tool list
Zep
None in its MCP tool list

Free plan

khive
No
mem0
Yes: 10,000 adds and 1,000 retrievals a month for 30 days, then 50,000 and 10,000 with your own LLM key
Zep
Yes: 10,000 credits a month

Paid plans

khive
$8, $25 and $100 a month
mem0
Usage-based billing past the free limits, then Custom. No public prices
Zep
$125 and $375 a month

Open source

khive
The khive runtime, Apache-2.0
mem0
The mem0 library, Apache-2.0
Zep
Graphiti, the graph engine, Apache-2.0. The Zep product itself is not open source

What each one stores

khive keeps four kinds of record in one workspace. A memory is one recollection, found again later by meaning. Entities (a service, a person, a document) and notes (a decision, an observation) are linked by edges that carry a relation such as depends_on, and a relation that does not fit the two kinds is refused. Tasks carry a priority, a due date, an assignee and the tasks they depend on. Knowledge atoms keep reference text as it was written. Tools lists every tool.

mem0 stores memories: text with metadata, scoped to a user, an agent, an app or a run, with MCP tools to add, search, list, update and delete them. Graph memory that links entities is listed under its Custom plan.

Zep keeps a temporal graph. You send it episodes (messages, text or JSON). It keeps the episodes, the entities in them and the facts between those entities, with summaries of a thread and of a user, and it supports custom entity and edge types. Its pricing page lists connectors for Google Drive, Slack, Notion, Gmail and Granola on Flex and above.

All three let an agent write to the store through MCP tools. Where they differ is in what happens next. mem0 and Zep work on what you send: mem0 stores extracted memories, and Zep builds its graph from the episodes. khive stores the record the agent wrote, with the kind and the relation the agent gave it, and adds a task list and reference documents in the same workspace.

Price

khive has three plans and no free tier: Start at $8 a month, Plus at $25 and Pro at $100, with annual billing 17% off. Plus monthly starts with a 14-day free trial, and the trial needs a card. Every plan reaches every tool. Plans differ in included credits and in limits, from 20,000 stored records on Start to 2,000,000 on Pro. A call is refused when credits run out, and nothing is billed afterwards. Plans has the details.

mem0 lists two plans, Free and Custom. For the first 30 days Free allows 10,000 adds and 1,000 retrievals a month, with mem0 paying for the model calls. With your own LLM key connected it allows 50,000 adds and 10,000 retrievals a month. Past the limits you can turn on usage-based billing with a monthly cap you set. No rate is published on its pricing pages. Custom has no public price.

Zep has a Free plan with 10,000 credits a month, then Flex at $125 a month and Flex Plus at $375 a month.

So both mem0 and Zep cost nothing to start, and khive does not. The prices also buy different things: mem0 and Zep run the extraction on their side, and khive stores what your agent writes. Past the free limits, khive's first step is a fixed $8 a month, Zep's is $125 a month, and mem0's is metered up to the cap you set.

Where khive is the weaker choice

Trying it costs money sooner. mem0 and Zep both have a free plan, and mem0's, with your own LLM key, allows more writes in a month (50,000) than khive's Start plan stores in total (20,000 records). khive has a 14-day trial on one plan, and the trial needs a card.

khive does not turn a conversation into memories for you. mem0 extracts memories from the text you send, and Zep keeps the episodes and builds entities and facts from them. With khive, what is in the store is what the agent chose to write.

khive has no connectors. Zep's paid plans list connectors that bring in documents and messages from Google Drive, Slack, Notion, Gmail and Granola. In khive, content arrives through the agent's tool calls.

khive does not sign users in through a company identity provider. Zep's MCP server signs members in through Google Workspace by default, and through a company's own OIDC provider on its Enterprise plan. khive signs you in with an emailed link or code, or with an API key.

khive cloud is hosted only, with no on-premises plan. mem0 lists on-premises deployment under Custom, and Zep lists deployment in your own cloud for enterprise customers. The khive runtime is open source under Apache-2.0, and running it yourself is a different thing from khive cloud.

Where khive is the stronger choice

Tasks live in the same store as memory and the graph. Neither mem0's nor Zep's MCP tool list has task tools. An agent on khive can record a decision, link it to the service it concerns and open a task for the follow-up, and the next session finds the task with its links.

The first paid step is small and fixed, at $8 a month. Zep's first paid plan is $125 a month, and mem0's next step after Free is metered.

Every khive plan reaches every tool, the graph tools included. mem0 lists graph memory under Custom.

Try it

Create an account at khive.ai/login, then follow the Quickstart.