Mem0 review, pricing and verdict
Memory layer for AI agents: an Apache-2.0 SDK, a managed Platform and an MCP server, with a free plan and 68.8k GitHub stars.
- Best for
- AI or backend engineer shipping production agents that need cross-session memory
- Not for
- Non-technical founder who wants a ready-made consumer memory app
Should you use Mem0?
Mem0 suits engineers who want drop-in agent memory with an open-source escape hatch: a free plan once an LLM key is connected, a quote-only tier for graph memory and SSO, and no self-serve paid plan. Treat its benchmark scores as vendor claims, because no third party has reproduced them.
One API and one open-source SDK that handle extraction, retrieval, graph links and consolidation, with the same code running self-hosted or on the managed Platform.
Community size and a broad plugin footprint create switching friction, but fact extraction is replicable and rivals such as Zep and Letta pitch different memory models.
Also for
- Platform engineer standardising MCP-based tooling across a team
- Backend and AI engineers building support chatbots that must remember users across sessions
Show 3 more
- Teams building agents on LangGraph, CrewAI or Strands Agents that need context beyond one conversation
- Developers who want one shared memory across Claude Code, Cursor and other MCP tools
- Platform teams weighing self-hosted against managed memory infrastructure
Also not for
- Team that needs memory to track how facts change over time out of the box
Main trade-off
Accuracy claims are self-reported and platform-only, and the advanced features sit behind a quote-only plan.
Mem0 pricing
| Plan | Monthly price | What it includes |
|---|---|---|
| Free | Free | Unlimited end users, Month one: 10,000 adds, 1,000 retrievals, model calls covered by Mem0, Own LLM key: 50,000 adds, 10,000 retrievals, Overage billed under a spend cap you choose, Community support |
| Custom | Custom | Everything in Free, Graph memory with entity linking, User profiles and Dream consolidation, SSO, audit logs, SLA, On-prem deployment, Limits sized to your volume |
How the pricing works, in full
Mem0 now sells two plans. Free has unlimited end users and, for the first 30 days, 10,000 adds and 1,000 retrievals a month with Mem0 paying for the model calls; once you connect your own LLM key the limits rise to 50,000 adds and 10,000 retrievals a month at no charge from Mem0, and usage past them is billed under a monthly cap you set. Custom is quote-only and adds graph memory, user profiles, Dream, advanced analytics, private Slack support, SSO, audit logs, an SLA and on-prem deployment. Starter, Growth and Pro are no longer sold to new accounts, and existing subscribers keep their terms.
Before you commit
- Memory-writing LLM calls bill to your own provider after the first 30 days, at about $3.55 to $5.34 per 1,000 adds on the models Mem0 recommends
- Usage past the free limits is billed per call under a monthly cap you set
What Mem0 does
Mem0 is an open-source memory layer for AI agents, founded in 2023, that stores facts from conversations and returns only what a turn needs. It offers an open-source SDK, a managed Platform, an MCP server and coding-tool plugins, and its April 2026 algorithm reports 92.5 on the LoCoMo benchmark.
Token-efficient memory algorithm
Single-pass, add-only extraction with semantic, keyword and entity retrieval fused into one ranking, which Mem0 says keeps each retrieval near 7,000 tokens against 25,000 or more for full context.
Temporal reasoning and memory decay
Resolves queries such as last week or as of March against memory timestamps, and ranks recently used memories higher; both shipped in May 2026.
Graph memory, profiles and Dream
Links facts through shared entities, keeps user profiles, and runs Dream in the background to merge duplicates and supersede outdated facts; all three sit on the Custom plan.
Bring your own key
Memories are written by your OpenAI, Anthropic, Azure OpenAI or OpenRouter model, and the ones Mem0 recommends run roughly $3.55 to $5.34 for every thousand adds on your own bill.
Show all 8 features
Open-source SDK and self-hosting
Python and TypeScript SDKs run fully self-hosted with your own vector store or against the managed Platform; the TypeScript SDK gained new providers and reranking in July 2026.
MCP server, OpenMemory and CLI
A live MCP server, the local-first OpenMemory and a CLI let Claude, Cursor, Windsurf and VS Code share one memory store without custom code.
Plugins and framework integrations
Ships plugins for Claude Code, Cursor, Codex, Antigravity, OpenCode, OpenClaw and Pi Agent, a native MemoryStore for Strands Agents, and n8n and Zapier nodes.
Webhooks and memory expiration
Webhooks announce memory changes to other systems, and SDK memory expiration, added on 27 June 2026, lets memories age out.
Works with
- LangGraph
- CrewAI
- Strands Agents
- Vercel AI SDK
- n8n
- Zapier
- Claude Code (plugin)
- Cursor (plugin)
- Codex (plugin)
- MCP clients
Missing integrations
- A self-serve paid tier for new accounts
Most users expect these and they are not currently supported.
Capabilities
- Memory layer
- Agent memory
- MCP
- Knowledge graph
- Vector search
- Open source
Strengths and limits
Strengths
- The core is open source and busy: the GitHub API showed 68,810 stars and a push on 6 November, and both SDKs shipped new releases in September, so a team can audit the code and self-host.
- The benchmark code is public, so a team can rerun Mem0's LoCoMo, LongMemEval and BEAM numbers itself instead of trusting a slide.
- The free plan is workable for a real prototype: with your own LLM key it allows unlimited end users and a monthly allowance of adds and retrievals at no charge from Mem0.
Limits
- After the first month the free plan needs your own LLM key, adds return an error until one is connected, and extraction cost lands on your provider account.
- Every published accuracy figure is Mem0's own, measured on the managed platform, and nobody outside the company has re-run them according to Atlan; open-source users should expect lower numbers.
- Graph memory, user profiles, Dream, SSO and on-prem deployment sit behind a quote-only Custom plan, and no self-serve paid tier is sold to new accounts.
- Its own llms.txt lags the pricing page, still listing a free tier of 1,000 memories, a Developer plan and SOC 2 Type II, while the site footer shows SOC 2 Type I and calls HIPAA and GDPR only Ready.
Mem0 alternatives
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| Mem0 This page | -- | Yes | Web | 3.6 |
 LettaPick Mem0 when you want a memory API that plugs into the agent stack you already run; pick Letta when you want an open-source agent platform in which the agent manages its own memory. | $20/mo | Yes | Web | -- |
 ExaSearch API that ranks the web by meaning for AI agents and retrieval pipelines, with Lite, Auto and Deep search modes, an asynchronous research Agent and an official MCP server. A free Starter plan with monthly credits needs no payment method. | $49/mo | Yes | Web | 4.0 |
| AlgoliaAI search and retrieval platform powering 1.75 trillion searches annually for 18,000+ businesses with semantic search, vector embeddings, and AI agents. | -- | Yes | Web · iOS · Android · Mac · Windows | 4.0 |
Mem0 facts
- Works offline
- Yes
- Minimum skill level
- Intermediate
- Price band
- Mid-range
- Maker founded
- 2023
- Training-data policy
- Not stated on the pages read; with a bring-your-own key the text you add is sent to your LLM provider under your account and its terms, and Mem0 stores the key encrypted.
- Compliance
- SOC 2 Type I; HIPAA Ready and GDPR Ready per the site footer
- Total funding
- $24M across a seed round (Kindred Ventures) and a Series A (Basis Set Ventures), announced 28 October 2025; Peak XV Partners, GitHub Fund and Y Combinator took part
Data and privacy
Apache 2.0 (open-source core); proprietary managed Platform for the hosted service
Questions about Mem0
What does Mem0 charge for the Platform?
Two plans are sold to new accounts: Free, and Custom by quote. Free costs nothing from Mem0 within monthly limits of 50,000 adds and 10,000 retrievals once you connect your own LLM key, then bills usage past those limits under a cap you set; Custom unlocks graph memory, user profiles, Dream, analytics, SSO and on-prem. Mem0's recommended models cost about $3.55 to $5.34 per thousand adds on your provider account, and the old Starter, Growth and Pro plans are no longer sold.
Is Mem0 really free?
Yes, in two ways. The open-source SDK is free to self-host, and the Platform's free plan covers unlimited end users, with Mem0 paying for the model calls during the first month. Accounts created before 26 October 2026 keep that window until 25 November 2026; after it, searches keep working but new adds fail until you connect a key.
Does Mem0 send my data to OpenAI or Anthropic?
With a bring-your-own key, yes: the text you add goes to your provider under your account and its terms. Mem0 stores the key encrypted and never shows it again, organisation owners connect it in Settings, and organisations created through the Vercel Marketplace cannot use it.
Is Mem0 or Zep better at remembering?
No head-to-head exists. An Atlan roundup refreshed on 9 October 2026 notes that Zep self-reports 90.2 on LongMemEval with GPT-5.4 while Mem0 self-reports 94.4 for its managed platform, from different papers, models and judges. Zep keeps a validity window for each fact, and Mem0 added time-aware retrieval in May 2026.
Can I use Mem0 inside Claude Code or Cursor?
Yes. The hosted MCP endpoint is mcp.mem0.ai, and a plugin shipped on 25 June 2026 shares one memory across Claude Code, Cursor, Codex and Antigravity. Its OpenMemory option keeps everything on your own machine through Docker.
About Mem0
Mem0 is a memory layer for AI agents and apps, founded in 2023 by Taranjeet Singh, its CEO, and Deshraj Yadav, and backed by Y Combinator. On 28 October 2025 it announced $24 million raised across a seed round led by Kindred Ventures and a Series A led by Basis Set Ventures, with Peak XV Partners, GitHub Fund and Y Combinator taking part. Its about page claims more than 200,000 developers. Mem0 sits between an application and its LLM calls, stores facts pulled from conversations and returns only what the current turn needs instead of replaying the whole chat history.
Read the full background
The April 2026 algorithm uses single-pass, add-only extraction, so no fact is overwritten, and scores semantic, keyword and entity signals in parallel before fusing them. Temporal reasoning and memory decay followed in May 2026. The managed Platform adds graph memory, user profiles and Dream, a background job that merges duplicates and supersedes outdated facts. The Apache-2.0 SDK installs with pip install mem0ai or npm install mem0ai and runs self-hosted or against the Platform, OpenMemory keeps a store local, and for coding tools there is a command-line client plus a remote MCP server. Mem0 reports 92.5 on LoCoMo, 94.4 on LongMemEval and 64.1 and 48.6 on BEAM at one and ten million tokens, at about 6,900 tokens per retrieval; the figures are its own, taken on the managed platform with a one-point margin, and the evaluation code is open source.
It fits backend and AI engineers who want drop-in memory for support bots, coding assistants and autonomous agents across sessions. Plugins and providers shipped through 2026 include a shared-memory plugin for Claude Code, Cursor, Codex and Antigravity on 25 June, a Vercel AI SDK provider, a native MemoryStore for Strands Agents on 24 August, and n8n and Zapier nodes, alongside LangGraph and CrewAI support. Checked directly on 11 November 2026, the GitHub API showed 68,810 stars, the Apache-2.0 licence and a push on 6 November, PyPI and npm both carried the 25 September releases (Python 2.2.1, Node 3.3.1), and the vendor's llms.txt still described a plan lineup its pricing page had dropped.
Look elsewhere when the memory should manage itself or track how facts change. Letta is an open-source agent platform from the MemGPT team in which the agent runs its own memory, and Zep, Cognee, Supermemory and LangMem are the other names roundups pair with Mem0; none of those four has a Hokai record yet. Atlan's comparison, updated on 9 October 2026, says every LongMemEval and LoCoMo score is a vendor measuring its own system and that no third party has reproduced Mem0's or Zep's figures. The free plan after the first month depends on connecting your own OpenAI, Anthropic, Azure OpenAI or OpenRouter key, and the memory-writing models Mem0 recommends include GPT-6 Luna, Claude Haiku 5.5 and DeepSeek V4.1 Flash.
On 26 October 2026 the pricing changed: new accounts get a free plan or a custom quote, and the older Starter, Growth and Pro plans stopped being sold. More retrieval and RAG infrastructure is on the AI search and retrieval infrastructure list, and context engineering explained, best AI search and retrieval infrastructure and best MCP servers cover where memory fits in an agent stack.
Sources and guides
Official sources
- Mem0 official site Vendor
- Official Documentation Docs
- GitHub Repository Docs
- GitHub repository GitHub
- Discord Community
Product facts on this page (pricing, plans, integrations, compliance) are read from Mem0's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.
Hokai guides that mention Mem0
Lists that include Mem0
More AI Tools on Hokai
LanceDBOpen-source embedded retrieval library with a distributed Enterprise lakehouse on the same Lance format; its GitHub repository shows 11.9k stars.
ReductoAgentic document platform that parses, extracts and splits PDFs, scans and spreadsheets for AI teams; its About page cites more than 5 billion pages processed.
ZeroxZerox is a free, MIT-licensed library that turns PDFs and 26+ document formats into clean markdown using vision models like GPT-4o, though development appears to have stalled.DeeplakeActiveloop's open-source GPU database for AI agents: Postgres interface, multimodal data lake, SOC 2 Type II certified, Apache-licensed core.
BrowserbaseHosted headless browsers for AI agents, with a free plan of 3 concurrent sessions, session replays and the open source Stagehand SDK.FirecrawlWeb data API for AI agents: search, scrape, crawl and interact, returning LLM-ready Markdown. Open source core with a free plan and an MCP server.