How to choose
First decide which job you have.
- You are building an app or agent that needs a memory of its users. Mem0, Letta, Zep, Cognee and Supermemory are mainly built for developers. Compare their SDKs, hosting, limits and price.
- You want one memory for a whole team and its AI tools. Ask who owns the memory, who approves changes, what record is kept, how access is set for each person, how agents connect, whether you can host it yourself or export it, and what the published price is.
This page judges each option on that second job.
At a glance
| Who approves changes | Access for each person | Hosting and export | Price as published | |
|---|---|---|---|---|
| Mem0 | A model adds memories automatically; no approval step is described | Reader and Owner roles on the hosted platform; SSO on Custom | Apache 2.0 open source or hosted; export on the hosted platform | Free and Custom; no price shown for Custom |
| Mensara | Governed areas wait for the owner; deleting or renaming always waits for an approver | Each person's agents work within that person's access | Ordinary files in a Git repository, with a private copy in your own GitHub | Not yet published; shown at checkout |
| Letta | Agents edit memory, including shared repositories several agents use; no approval step from a person is described | Admin, Editor and Analyst roles; SSO and RBAC on Enterprise | Apache-2.0, run locally or self-host a server | Free $0, Pro $20 a month, Teams Pro $20 a seat a month, Enterprise custom |
| Zep | Writes are switched on or off for each connection; no approval of each change is described | On Enterprise, roles for people and attribute-based access for agents' API keys | Graphiti is open source; your own cloud on Enterprise | Free (10,000 credits a month), Flex $125 a month, Flex Plus $375 a month, Enterprise custom |
| Cognee | A model extracts facts; conflicts are kept and can be settled later | Read, write, delete and share permissions on each dataset | Apache-2.0, self-host with Docker; export as a COGX archive | Open source free; Cloud Standard $0 a month plus $0.10 per 1M tokens (one-time $10 starting credit), Premium $99 a month plus tokens, Enterprise by contact |
| Supermemory | A memory model decides what to keep and forget; no approval by an owner or another person is described | Members can be limited to certain namespaces on the Scale plan | MIT-licensed GitHub repository; free local binary; self-hosted option on Scale; export as JSON from Settings | Free $0, Pro $19, Max $100, Scale $399 a month, Enterprise custom |
Mem0
In its own words, "the memory layer for AI agents and applications". Best for developers adding memory of each user to an app or agent.
For team use: by default a model reads messages, pulls out facts and adds them, and the documentation describes no step where a person approves a memory before it is saved. Organizations and projects on the hosted platform have two roles, Reader and Owner. Each memory keeps its change history, and the hosted platform adds a project-wide event feed. SSO, audit logs and on-premises deployment are on the Custom plan. Mem0 states "SOC 2 (Type 1), HIPAA, and GDPR compliant".
Drawback for a team: it is a memory layer for an app, not one memory a company's people share and approve, and the governance features sit on a plan with no published price. See Mensara vs Mem0.
Mensara
One shared memory, a Mind, for a team and their AI agents. Best for founders and small teams who want the people and AI tools across the team to work from the same records.
For team use: people and their agents propose updates where the work happens. Some permitted changes save after an automatic check; changes to governed areas, such as decisions, wait for the owner's approval, and deleting or renaming a file always waits for an approver. An area can give someone permission to suggest edits without giving them authority to approve. Each person's agents work within that person's access. The Mind keeps a signed, tamper-evident receipt for each read and change, including refusals. Agents connect from Claude, Codex, Cursor and ChatGPT on the web (developer mode) over MCP. Saved records and their version history are ordinary files in a Git repository that your team can copy and keep.
Drawback for a team: public pricing is not yet published, and its public pages list no SOC 2 or HIPAA certification. It is not a memory API for building your own app, and approvals in governed areas take an owner's time.
Letta
"The platform for building stateful agents." Best for developers building agents that learn and keep their own memory, and teams that want a shared AI coworker.
For team use: agents edit their memory in a git-backed memory filesystem, and several agents can share the same memory files. Letta's docs describe an optional review step in memory consolidation that does not ask you for approval. Organizations have Admin, Editor and Analyst roles; agents are private by default and can be shared with the organization. SSO and role-based access control are on Enterprise. Letta is open source under Apache-2.0 and can run locally or on a self-hosted server.
Drawback for a team: memory is edited by agents, including shared repositories several agents commit to, with no approval step from a person described, and no SOC 2 or HIPAA claim was found on its site.
Zep
"The enterprise context layer for agents." Best for companies building production agents on knowledge graphs.
For team use: on the Enterprise plan, people get account- and project-scoped roles, and agents' API keys get attribute-based access. Its Context MCP server lets Claude, ChatGPT, Claude Code, Codex, Cursor and others connect, with each person signing in and reaching their own user graph. Agents write when a connection allows it; writes are switched on or off for each connection, with an account-level kill switch, and no approval of each change by a person is described. Graphiti, Zep's temporal knowledge-graph framework, is open source.
Drawback for a team: Zep states SOC 2 Type II, but SSO, audit logs, bring-your-own-key, running in your own cloud and a HIPAA BAA are on the Enterprise plan, and no approval of each change by a person is described.
Cognee
"Memory for Agents, Teams, and People." Best for teams that want an open-source memory and knowledge-graph pipeline they run and configure themselves.
For team use: a model extracts entities and relationships from what you add. When facts conflict, Cognee keeps both and records which one is current, and a person can settle the conflict. An opt-in provenance ledger records when, by whom and in which run each document was stored. Datasets carry read, write, delete and share permissions within tenants and roles. Integrations and an MCP server let Claude Code, Cursor, Codex and other agents read and write memory. It is open source under Apache-2.0 and self-hosts with Docker.
Drawback for a team: changes go in first and conflicts are reviewed afterwards, governance is permissions you configure on datasets, SSO is offered only on some Enterprise engagements, and its SOC 2 audit is in progress.
Supermemory
"A memory API for AI agents and apps." Best for developers who want a hosted memory engine for their agents and users.
For team use: a memory model decides what to learn, what is important and when to forget, and no approval by an owner or another person is described; the person saving through its MCP server can review a draft first. Namespaces are an authorization boundary, so on the Scale plan an organization member can be limited to certain namespaces, and API keys can be scoped or read-only. Its MCP server works with Claude Desktop, Claude Code, Cursor, ChatGPT and others. Organization owners or admins can export up to 25,000 documents and memories as JSON from Settings. Its GitHub repository is MIT-licensed, and its pricing page lists a free local binary for up to 10,000 documents. The Scale plan lists SOC 2, a HIPAA BAA and a self-hosted option.
Drawback for a team: the memory model, not a person, decides what is kept, and SSO is on the Enterprise plan.
Questions
What is the best Mem0 alternative for a team?
It depends on the job. For an app or agent that needs a memory of its users, compare Letta, Zep, Cognee and Supermemory with Mem0. For one memory that a team and its AI tools share, with approval for governed areas and a record of activity, look at Mensara.
Which of these let a person approve a change before it is saved?
On the pages checked, Mensara is the one that describes an owner or approver doing it: changes to governed areas wait for the owner's approval, and deleting or renaming a file always waits for an approver. The others describe memory written by a model or an agent, with controls such as per-connection write switches in Zep and conflict review after the fact in Cognee.
Which can we host ourselves?
Mem0, Letta and Cognee are open source under Apache 2.0 and can run on your own infrastructure. Zep runs in your own cloud on its Enterprise plan, and its Graphiti framework is open source. Supermemory's GitHub repository is MIT-licensed, it offers a free local binary for up to 10,000 documents, and its Scale plan lists a self-hosted option. For Mensara, your Mind's content lives on its own server, and who runs that server depends on the hosting arrangement you choose.
Which publish a price?
Letta, Zep, Cognee and Supermemory publish prices on their own sites. Mem0 shows a Free plan with stated limits and a Custom plan with no price. Mensara's public pricing is not yet published; the price, billing period, tax and terms are shown at checkout.
Sources
All competitor facts checked 18 November 2026 on each company's own site; Zep, Cognee and Supermemory facts rechecked 19 November 2026.
- Mem0: home page, pricing, platform vs open source, adding memories, organizations and projects, GitHub
- Letta: docs, shared memory, pricing, memory, team permissions, self-hosting, GitHub
- Zep: home page, pricing, governance, Context MCP server, agent memory over MCP, Graphiti
- Cognee: home page, pricing, permissions, memory versioning, GitHub
- Supermemory: home page, pricing, how it works, namespaces, ChatGPT connection, data export, GitHub
- Mensara: About, Changes, approvals and receipts, Backup, restore and ownership, Ownership and data flow, Account and billing
Related
Practical guides for people and agents