Mem0
Drop-in memory layer for AI agents and apps
About Mem0
Mem0 is a memory layer for AI agents. The problem it addresses is narrow and familiar: a model has no memory between sessions, so every conversation either starts from nothing or drags the entire history along as context, which is slow and expensive. Mem0 sits between your app and the model. You hand it conversation turns, it extracts the facts worth keeping, and later you query it for the ones relevant to the current request instead of replaying everything that came before.
The open source core is Apache 2.0 licensed and can be self-hosted, and there is a managed platform on top of it with a free tier. It is Y Combinator backed. The project is unusually active for its age, with more than 2,600 commits and plugins for Claude Code, Codex, Cursor and Kimi Code that wire memory into an existing coding agent rather than requiring you to build around it. One thing to be clear-eyed about: the free Hobby tier allows 10,000 add requests and only 1,000 retrieval requests a month, and retrievals are what a working application does constantly. Anything past a prototype lands on a paid tier quickly, and the jump from $19 to $249 is where graph memory and the Dream consolidation feature live. Mem0's marketing claims lower token costs and faster responses; those numbers are not published on the pricing or home page, so treat them as a hypothesis to test against your own workload rather than a specification.
You install the SDK for Python or Node and create a client with an API key, or point it at a self-hosted instance. After a conversation turn, you call add with the messages and a user id, and Mem0 extracts and stores what is worth remembering rather than saving the transcript verbatim. Later, you call search with a query and a filter such as a user id, and it returns the relevant memories to inject into your prompt. Existing memories are updated rather than duplicated as new information arrives, so contradictory facts converge instead of accumulating. In practice the loop is three steps: add anything, let Mem0 learn, retrieve on the next interaction. If you would rather not touch the code, the editor and agent plugins add the same behaviour to Claude Code, Codex, Cursor and Kimi directly.
- •Fact Extraction, Not Transcript Storage - Mem0 condenses chat history into compact memories rather than keeping raw logs, which is what makes retrieval cheap enough to do on every turn
- •Python and TypeScript SDKs - A small API surface, mainly add and search with filters, so wiring it into an existing agent is a couple of calls rather than a pipeline change
- •Scoped Memories - Memories are scoped by user id and project, with unlimited end users on every tier including the free one
- •Graph Memory - Entity linking across stored memories, available on the Pro tier and above
- •Dream Memory Consolidation - Mem0's newer approach to keeping memory accurate as the store grows, also a Pro tier feature
- •Agent and Editor Plugins - First-party plugins for Claude Code, Codex, Cursor and Kimi Code, so a coding agent gains persistent memory without any integration work
- •Apache 2.0 and Self-Hostable - The open source core carries no usage restrictions, and Enterprise adds an on-prem deployment option
Teams building assistants, support agents or any product where a user comes back and expects to be remembered, and who do not want to build fact extraction, deduplication and retrieval themselves. It is also useful to developers who simply want their coding agent to stop forgetting project conventions between sessions, which the plugins handle without code. Self-hosters get the whole core under Apache 2.0. It is not for you if your application is genuinely stateless, or if a vector store you already run plus a bit of prompt assembly covers your needs, because Mem0 is another service to operate for a problem you may already have solved. Be careful with the free tier if you are evaluating under load: 1,000 retrievals a month is a demo budget, not a pilot budget.
Pricing
$0 - $249/mo
- Free$0/mo
- Starter$19/mo
- Pro$249/mo
- EnterpriseContact sales
From the vendor pricing page, 2026-10-12
