MemoryLake
THREE LAYERS, ONE MEMORY

Memory that travels the whole way.

MemoryLake sits between the systems your work already lives in and the people who need answers from it. Data comes in, models and agents reason over it, and the result lands where your team is already talking.

One pipeline, three layers

Each layer is useful on its own. Together they close the loop: nothing your team learns has to be explained twice.

1

Connect your data sources

2

Give models and agents the memory

3

Put it where your team is talking

1LAYER 1 · INGEST

Connect your data sources

Memory has to come from somewhere. Point MemoryLake at the drives, suites and conversations your team already uses, and it keeps reading them as they change.

Cloud drives & suites

Lark
DingTalk
WPS 365
OneDrive
SharePoint
Google Drive
Dropbox
Baidu Netdisk

Connectors bind to a personal space. Team and organization spaces ingest through upload, API and SDK instead.

Conversation history

ChatGPT
Claude
Gemini

Direct ingest

Upload
REST API
SDK
2LAYER 2 · REASON

Give models and agents the memory

The memory is model-agnostic. Reach it over MCP from a coding harness, from a desktop chat app, from your own code, or from an autonomous agent — the same memory, whichever one is holding it.

Coding harnesses

Codex
Cursor

Chat apps & extensions

Claude
Gemini
Chrome extension

Autonomous agents

Protocols & SDKs

REST API
CLI
Memory Routerprivate preview

Model-agnostic by design

Your memory is not tied to whichever model wrote it. Switch models and the memory comes along.

OpenAIAnthropicGeminiQwenDeepSeek
3LAYER 3 · DELIVER

Put it where your team is talking

An answer nobody sees is not an answer. Connect a messaging channel and the agent picks up mentions, files and threads in the chat your team is already in — no new tool to adopt.

Messaging channels

Slack
Roster

Or hire the whole employee

Roster goes one step further than a channel: a digital employee with its own account in your IM, a job description, quarterly OKRs, long-term memory in MemoryLake, and a manager who approves what matters. It runs inside DingTalk today; the sending layer is pluggable, so other platforms can follow.

Meet Roster

Step-by-step setup guides

Official walkthroughs for the integrations people set up most often.

Something missing from the list?

The registry grows with what customers actually connect. Tell us what you need and we will tell you where it sits on the list.