MemoryLake
AI MEMORY · Neutral by design. Works with every AI and agent.

The memory lake for every AI.

Multimodal memory — conversations, documents, spreadsheets, audio, video. Productivity intelligence for individuals and enterprises.

Get Your Unlimited Memory

Free forever · Private · Encrypted · Portable

Your memory works with
ChatGPTClaudeQwenGeminiOpenClawAutoGPTManusPerplexity+ any API

01 / What makes MemoryLake different

What can MemoryLake do that your AI can't?
14 things, side by side.

MemoryLake keeps typed, versioned memory instead of a chat log — that is what makes these 14 answers possible. Tap any question: both AIs see the same one.

E
Ethan's Memory Passport
PrivateEncryptedPortable
13,000
Entities
327
Agents
410K
Messages
1,021
Excel
766
PDF
30
Audio
47
Video
199
Images
93
Skills
8
Open Data
2024.1
Since
AI
AI without MemoryLake
ChatGPT · Claude · any LLM
What's in my memory?
What's in my memory?
ML
AI with Your MemoryLake
OpenClaw · D1 · your memory
What's in my memory?
What's in my memory?

Seeing something that changes how you think about AI?

Get Your Unlimited Memory

Free to use · Premium plans available

02 / What lives in your passport

What kinds of memory does it store?
Six, and all of them yours.

Background, fact, event, dialogue, reflection and skill — six typed memories, stored separately so retrieval targets the right kind instead of searching one flat blob.

Background Memory

The things that never change — your values, your world model, manually set by you. Read-only.

Fact Memory

Everything verifiable — checked for conflicts, versioned, traceable to its source.

Event Memory

Your timeline — what happened, when, in sequence. Builds your life narrative.

Dialogue Memory

Every conversation, compressed and searchable. Nothing ever lost.

Reflection Memory

Deep patterns your AI noticed. How you think, how you decide.

Skill Memory

Methods you built once — reused forever, across any AI, any session.

03 / Performance

What does memory actually buy you?

A 10,000× larger working set, 99.8% recall accuracy, 91% fewer tokens and 97% lower latency — what that looks like in production.

0K x

Data Scale Amplification

Support 10,000x more data compared to direct LLM file reading

0.0%

Memory Recall

Production-tested on 100M+ complex mixed documents

0.0%

Token Cost Reduction

Compared to direct LLM file reading, with millisecond latency

0.0%

Latency Reduction

Compared to direct LLM file reading

04 / #1 Globally

How does MemoryLake score on LoCoMo?

94.03% overall — the highest score in this evaluation. LoCoMo is the long-conversation memory benchmark introduced by Maharana et al. (2024): multi-session dialogues with temporally distributed Q&A. The chart breaks the score down by single-hop, multi-hop, temporal and open-domain recall, and it is cross-session integration that separates the systems.

Read moreRead the LoCoMo paper (arXiv)
MemoryLake
94.03%
Benchmark 1
92.32%
Benchmark 2
91.21%
Benchmark 3
85.22%
Benchmark 4
80.76%
Benchmark 5
66.67%
Benchmark 6
64.2%

05 / Your data. Your rules.

Can we read your memory?
No. We architecturally cannot.
Everything belongs to you.

Security is the origin of our design, not a layer on top: you own the memory, you decide who sees what, and you can delete all of it.

Not "we promise we won't look." We architecturally cannot look. Triple-party encryption means no single party — including us — ever holds all the keys.

Your three rights

You own your memory.

Export everything, anytime. One click. No questions asked.

You control who sees what.

Each AI sees only what you explicitly allow. Not a setting. Not a promise. Architecture.

You can delete everything.

Permanent. Irreversible. We keep no copies. No backups. Gone.

If your security team asks:

ISO/IEC27001CERTIFIEDISO/IEC 27001Information security management
SERVICE ORGSOC 2TYPE 2SOC 2 Type 2Security & availability audit
ISO/IEC20000-1CERTIFIEDISO/IEC 20000-1IT service management
EUGDPRCOMPLIANTGDPREU personal data protection
PCIDSSCOMPLIANTPCI DSSPayment card data security
ISO9001CERTIFIEDISO 9001Quality management
ISO14001CERTIFIEDISO 14001Environmental management
ISO45001CERTIFIEDISO 45001Occupational health & safety

Certificate numbers and audit reports are available on request — ask your account contact.

06 / Open Data

What public data is built in?

Seven public corpora are indexed and queryable out of the box — papers, filings, trials, markets, compounds, patents and economic series. No ETL, no API keys, no setup.

40M+ papers

Academic Papers

PubMed, arXiv, bioRxiv

3M+ filings

SEC Filings

EDGAR

500K+ trials

Clinical Trials

ClinicalTrials.gov

Live feeds

Financial Data

Real-time stocks, crypto, forex

2M+ compounds

Drug Databases

ChEMBL, DrugBank, FDA

10M+ US patents

Patent Search

USPTO

Global coverage

Economic Data

FRED, BLS, World Bank

07 / Conflict Detection

What happens when two memories contradict each other?

MemoryLake catches the contradiction the moment the memory is ingested, resolves it by a strategy you choose, and keeps both versions in the audit trail.

Source A — ChatGPT

"User prefers dark mode UI"

Updated 2 hours ago

Conflict Detected
Source B — Claude

"User prefers light mode UI"

Updated 5 hours ago

Auto-Resolved

Most recent source wins. Previous value preserved in version history with full provenance chain.

Real-Time Detection

Conflicts are identified the moment new memories are ingested, before they can propagate through your system.

Smart Resolution Strategies

Choose from recency-based, source-priority, confidence-weighted, or manual resolution — or define your own rules.

Full Audit Trail

Every conflict, every resolution, every version change is permanently recorded with timestamps and source attribution.

Cross-Source Validation

Automatically cross-references memories across multiple AI sessions and data sources to ensure consistency.

08 / Memory Provenance

Where did this memory come from?

Every memory carries its source, creation time, transformation history and access log — a provenance chain you can export for an audit.

a3f8c2d09:14 AMPDF Upload

Document ingested: Q4_Financial_Report.pdf

Extracted 47 facts, 12 entities, 3 temporal events

b1e9f4a09:14 AMLakeBuilder

Memory graph updated with 47 new nodes

Vector embeddings generated, graph connections established

c7d2a1b09:15 AMConflict Engine

Revenue figure conflict detected with Q3 projection

Q3 projected $4.2M vs Q4 actual $3.8M — flagged for review

d4f6e3c09:15 AMAuto-Resolver

Conflict resolved: Q4 actual supersedes Q3 projection

Resolution policy: actual > projection. Version v3 → v4

e2a8b5d09:16 AMQueryAgent

Memory served to Claude session #8842

Context: "What was our Q4 revenue?" — 23ms latency

Git-like Versioning

  • Full commit history for every memory change
  • Branch and merge memory timelines
  • Roll back to any previous state
  • Diff view between memory versions
  • Immutable audit log

Source Attribution

  • Trace any fact to its original source
  • Cross-reference across AI sessions
  • Cryptographic integrity verification
  • Compliance-ready provenance chain
  • Export full audit reports

09 / Features

What makes it enterprise-grade?

Git-like versioning, conflict detection, distributed processing, the MemoryLake-D1 parsing engine and RL-based analysis — under ISO 27001, SOC 2, GDPR and CCPA with end-to-end encryption.

Git-like Memory Versioning

Track every change with branches, commits, and time travel capabilities.

Conflict Detection

Detect logical conflicts, implicit knowledge contradictions, and hallucination conflicts.

Distributed Computing

Scale memory operations across clusters with Python SDK.

MemoryLake-D1 Engine

Visual + logical dual validation for parsing complex Excel, PDF layouts.

RL-Based Analysis

Reinforcement learning for automatic data understanding and insight generation.

Enterprise Security

ISO27001, SOC2, GDPR, CCPA certified with full audit trails.

10 / Data Sources

What can MemoryLake ingest?

Documents, spreadsheets, images, audio and video, plus enterprise systems, APIs and online docs — 20+ integrations, parsed by our own MemoryLake-D1 vision-language model.

Multimodal Files

Conversations
Text
Documents
Images
Video
Audio

Enterprise Data

MySQL
PostgreSQL
Delta Lake
Hudi
Iceberg

APIs & Protocols

REST API
MCP

Online Documents

Office365
Google Workspace
WPS365
Dropbox
Lark
Dingtalk

11 / Scale

How much does it handle today?

1.5M+ users and 15,000+ enterprises, 130M+ memory requests, 10M+ files processed and 50M+ memory code runs.

0M+Users
0K+Enterprises
0M+Memory Requests
0M+Files Processed
0M+Memory Code Runs

12 / Integrations

Where can you use it?

In the tools you already have. Install a plugin and your agent syncs itself, pick a platform that ships memory built in, put an agent in your team's chat, or call the protocol yourself — one memory tree, four ways in.

Plugin

Coding agents

Install once. The agent recalls what it needs before it reads and writes memory back when something changes — you never have to ask it to.

Built in

Agent platforms

Memory is already wired into the platform. Connect an API key and every agent you run there inherits it.

Team chat

Where your team already talks

Add an agent to a direct message or a group channel. The thread itself becomes memory your other tools can read.

Protocol

Call it yourself

MCP for on-demand lookups from any client, the CLI for terminals and CI, REST for your own code.

Every route reads and writes the same memory. Change tools, keep the context.

See every integration

Ready for one memory
across every AI?
Yours forever.

One Memory Passport carries your memory across every AI you use — encrypted, portable, and yours to keep.

Start free. Keep what matters.

Upgrade when you're ready.

Get Started

Free to use · No credit card · Premium plans available