Grow
Build a domain one leaf at a time
Use Notes to shape a tree that fits the way the domain is actually organized. Leaves hold the detailed notes and attachments that become the working source material.
About
MDS is a workspace for domains of knowledge.
You grow a tree of notes and attachments, inspect the material, and talk to the domain. The agent searches that tree and shows the ground under the answer. When it uses broader model knowledge, you can pull the useful part back onto a leaf.
“Imagine you had access to the knowledge GPT speaks from. That you could look into it, edit it, tell it what to remember — and take part in building it. Not the whole world’s knowledge. Precisely the domain that matters in your work or your hobby. You would be steering part of its memory. Context would no longer be just the chat history. It would contain the memory you had built. And the agent would speak from that understanding of the domain.”
Generate
Draft trees, branches, and leaves faster
From Notes, use generation to draft a full tree, extend a branch with more branches, or generate leaf notes so the domain does not have to be built entirely by hand one item at a time.
Review
Activate drafting and review where it matters
A tree can opt into its own review workflow when the domain needs more control. That lets people draft material first, submit it for review, and approve or reject it inside the same workspace, while trees that do not need that process can stay lightweight.
Inspect
See what the material actually says
Use Search to inspect notes and attachments directly, review highlighted matches, and jump back into the exact branch or leaf that produced them.
Talk
Ask the domain, not a general model
Use Agent to ask questions against the trees you built. The answer is shaped by the material in that domain instead of relying only on broad model knowledge.
Grounding
See the ground under the answer
When the agent uses material from the tree, it shows the grounding so you can inspect the path back to the source and judge whether the answer is well supported.
Broader answers
Use outside knowledge deliberately
When the agent steps outside the local tree and uses broader model knowledge, that path stays explicit instead of pretending everything was grounded in your material.
Capture
Pull useful answers back onto a leaf
When a broader answer contains something worth keeping, you can turn that result into structured knowledge by adding it back into the right place in the tree.
Collective memory
Hold the domain together
The agent has your understanding of the domain — the tree you built together — so it can recognize the connections in that material and speak from them. Context is not the chat history. It is a shared memory.
Extensibility
Add tools, and the agent grows
Give the agent a tool for a particular context, and it can use that capability there. Then it does not only understand the domain. It can act in it. Each new tool extends what it can do, and the agent grows over the capabilities you have given it.
Safety
Restore before permanent removal
Deleting a tree, branch, or attachment does not erase it immediately. The workspace keeps a 7-day recovery window first, so accidental deletes can be restored before final cleanup.
Access
Manage access through the groups teams already use
MDS does not need a separate user list inside the workspace. People sign in with Microsoft identity, and access can follow the Entra groups a team already uses.
Working surfaces