MemoryLake: Multimodal AI Memory and the Memory Passport
MemoryLake is an AI memory platform that stores what you and your AI tools learn, from conversations and notes to PDFs, spreadsheets, audio and video, and serves the right pieces back to whichever AI you’re using. Individuals get a “Memory Passport” that carries their context between ChatGPT, Claude, Claude Code and agent platforms like OpenClaw. Companies get shared memory with permissions, version history and audit trails. Its distinguishing ideas are Git-style versioning for memories, automatic handling of contradictions, and its own model for reading complicated documents.
The quick answer: MemoryLake aims to be a neutral memory layer that belongs to you rather than to any one AI company, and it’s unusually broad: personal memory, coding-agent memory, a drop-in proxy for developers and an enterprise product. It launched in early 2026 from a team led by a former Alibaba data-infrastructure architect. Its headline numbers are self-reported, and we found no independent reviews yet. A disclosure: we make Carly, an AI assistant with its own memory and workspace, so we compete on the memory side. Everything below comes from MemoryLake’s site, terms and pricing, and Chinese tech press coverage.
MemoryLake at a glance
Checked against MemoryLake’s site, pricing and terms of service on November 12, 2026.
| MemoryLake | |
|---|---|
| What it is | A multimodal memory platform for people, AI agents and companies |
| Memory types | Background, fact, event, dialogue, reflection and skill |
| Works with | ChatGPT, Claude, Claude Code, OpenClaw, Hermes Agent and any MCP client; REST, CLI and SDK |
| Distinct features | Git-style memory versioning, conflict detection, built-in public datasets |
| Pricing | Free plan; Pro $19 a month; Premium $199 a month; pay-as-you-go credits |
| Security claims | ISO 27001, SOC 2, GDPR and CCPA; reports on request |
| Company | DataCloud Technology Pte. Ltd., Singapore; R&D and operations in Japan |
What MemoryLake does
MemoryLake’s homepage opens with “Neutral by design. Works with every AI and agent.” Its about page explains the philosophy: “AI can be replaced, but your memory and decision-making ability must be preserved.” The idea is that models are interchangeable, and what makes an AI useful to you is the memory it has of you and your work, which should stay yours when you switch models.
MemoryLake sorts memory into six types, stored separately so a question goes to the right kind:
| Type | What it holds |
|---|---|
| Background | Things that rarely change, like your values, set by you and read-only to the AI |
| Fact | Verifiable information, checked for conflicts, versioned and traced to its source |
| Event | Your timeline: what happened and when |
| Dialogue | Past conversations, compressed and searchable |
| Reflection | Patterns the AI has noticed in how you think and decide |
| Skill | Methods you’ve worked out once and want reused |
Versioning and contradictions
Most memory tools quietly overwrite or pile up facts. MemoryLake treats memory more like code. Every change gets a commit, and you can view the differences between versions, roll back to an earlier state, or branch and merge memory timelines.
When a new memory contradicts an old one, such as “User prefers dark mode UI” followed later by “User prefers light mode UI,” MemoryLake flags it the moment it’s stored and resolves it by a strategy you choose: most recent wins, a trusted source wins, the more confident one wins, or a person decides. Both versions stay in the audit trail. Every memory also carries its source, creation time and access log, which MemoryLake pitches to companies that need to show where an AI’s answer came from.
Built for messy documents
MemoryLake reads more than chat. It ingests documents, spreadsheets, images, audio and video, plus connected drives and enterprise systems, and parses them with its own vision-language model, MemoryLake-D1, designed for complex Excel layouts, scanned PDFs, flowcharts and meeting recordings. It also comes with seven public datasets already indexed, covering research papers (PubMed, arXiv and bioRxiv), company filings, clinical trials, live market data, chemical compounds, patents and economic statistics, so an agent can draw on them without any setup.
Ways to use it
MemoryLake offers several ways in, all reading and writing the same memory:
- Memory Passport. A personal memory that follows you between ChatGPT, Claude and other AI apps, with controls over what’s stored, shared or forgotten. It’s free for individual use.
- Coding agents. A Claude Code integration over MCP loads project decisions and conventions at the start of each session, free for individual developers. OpenClaw and Hermes Agent have it built in with an API key.
- MemoryLake Router. A proxy developers put between their app and the OpenAI, Anthropic or Google APIs. Changing one URL adds long-term memory and trims redundant history from each request.
- Team chat and Roster. An agent can join channels and direct messages, and Roster goes further with what MemoryLake calls “a digital employee”: an agent with its own account, a job description and long-term memory, working in Feishu, WeCom, DingTalk, Slack and Telegram.
- WorkBrain. The enterprise product, with shared company knowledge, company-specific terms and rules, and permissions down to individual table rows.
Privacy and security
MemoryLake makes strong privacy claims. Its homepage says “We architecturally cannot look,” explaining that its encryption is split among three parties so that no single one, including MemoryLake, holds all the keys. It promises one-click export, per-AI access controls, and permanent deletion with no copies or backups kept. It lists ISO 27001, SOC 2, GDPR and CCPA among its certifications and says certificate numbers and audit reports are available on request.
Benchmarks and scale claims
MemoryLake reports a score of 94.03% on LoCoMo, a benchmark of memory across long, multi-session conversations, which it describes as the highest in its evaluation. It also claims 99.8% recall accuracy and up to 91% fewer tokens than having a model read files directly. These are company figures. LoCoMo results are self-reported across the memory industry with differing methods; Cognee, a rival, has published a critique of how such scores get inflated. Compare carefully before relying on any one number.
The site also cites more than 1.5 million users and 15,000 enterprises. According to QbitAI, a Chinese tech outlet, the team built MemoryLake on lessons from Powerdrill, a data-analysis agent it launched in 2024 that had more than 1.5 million professional users overseas.
Pricing
MemoryLake meters everything by tokens, whether it’s reading a PDF page, storing a memory or answering a query:
| Plan | Price | Monthly tokens |
|---|---|---|
| Free | $0 | 300,000 |
| Pro | $19 a month | 6.2 million |
| Premium | $199 a month | 66 million |
| Credits | About $3.125 per million tokens | Pay as you go, no expiry |
Yearly billing saves about 17%. MemoryLake’s own estimates translate the Free plan to roughly 40 document pages or 1,200 memory searches a month.
The company
MemoryLake’s terms of service name DataCloud Technology Pte. Ltd., a Singapore company, as the operator, and its about page lists research, development and operations in Japan. Chinese tech press describes the team behind it as Zhibian Technology, founded by Zhan Chaoqun. In an interview with TMTPost, Zhan said he spent 11 years at Alibaba, where he led the team behind its cloud-native data warehouse. The same team makes Powerdrill and a multimodal data platform called Relyt. QbitAI reported MemoryLake’s official launch in February 2026. We couldn’t confirm the company’s funding from a primary source.
How MemoryLake compares
MemoryLake competes with memory layers like Supermemory, Honcho, Zep, Cognee, LangMem, Redis Agent Memory and Mem0. It stands apart on breadth: personal, developer and enterprise products in one platform, multimodal parsing with its own model, Git-style versioning and preloaded public datasets. Zep is its closest match on tracking how facts change over time, and Cognee on open-source flexibility, which MemoryLake doesn’t offer.
If you want a personal assistant with memory already in it rather than a memory layer to wire up, Carly has memory built in: it saves named memory documents that it reads and searches in later conversations, searches a knowledge base of files you upload by meaning, and keeps your to-dos, data tables and CRM in one workspace. It’s simple to start, works the same inside ChatGPT, inside Claude or by text, and also labels and sorts email across all your accounts, handles advanced scheduling, and makes and answers calls and texts.
Who MemoryLake suits
MemoryLake is a good fit if:
- You switch between AI tools and want one memory that follows you.
- Your company’s knowledge lives in spreadsheets, scanned PDFs and recordings, not just text.
- You need version history, contradiction handling and audit trails for AI memory.
- You want memory for Claude Code or OpenClaw agents without building it.
Look elsewhere if:
- You want open-source software you can run and inspect yourself.
- You need independent benchmarks and reviews before choosing a vendor.
- You need a vendor with a long operating history; MemoryLake launched in 2026.
FAQ
What is MemoryLake?
MemoryLake is a multimodal AI memory platform. It stores conversations, documents, spreadsheets, audio and video as typed, versioned memory and serves it to ChatGPT, Claude, coding agents and other AI tools.
What is the MemoryLake Memory Passport?
The Memory Passport is MemoryLake’s personal memory product. It carries your context and preferences between AI apps like ChatGPT and Claude, and it’s free for individual use.
Is MemoryLake free?
MemoryLake has a free plan with 300,000 tokens a month. Paid plans are Pro at $19 a month and Premium at $199 a month, with pay-as-you-go credits beyond that.
Does MemoryLake work with Claude Code?
Yes. MemoryLake connects to Claude Code as an MCP server, loading project memory at the start of each session. It’s free for individual developers.
Who makes MemoryLake?
MemoryLake is operated by DataCloud Technology Pte. Ltd. in Singapore. Chinese press reports the team is led by Zhan Chaoqun, a former Alibaba data-infrastructure leader, and also makes the data-analysis agent Powerdrill.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
