> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reducto.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Document server

> One live copy of each .docx that editor tabs, agents and your backend edit together.

The document server holds the one copy of each document that everyone edits. Browser tabs and agents send their edits to it, and it sends each edit it accepts to every subscriber. Agents can also send code, which runs on the server's copy. You run it in your own cloud.

> [!NOTE]
>
> The document server is available on request. It holds Word documents (.docx) only today. Support for PowerPoint (.pptx) and Excel (.xlsx) is coming.

> [card] [Request access to the document server](https://reducto.ai/contact?source=docs_office_document_server)
>
> Tell us about your deployment and we will set you up with an image and the setup guide.

```mermaid theme={null}
flowchart LR
  tabA[Editor tab] <-->|edits| server[(Document server)]
  tabB[Editor tab] <-->|edits| server
  agent[Agent] -->|code to run on the document| server
  backend[Your backend] -->|create, tokens, download| server
  server --> store[(Your bucket: .docx and edit log)]
```

## Why one source of truth

Without a server, each writer has its own copy of the document: the user's tab, a second tab, the agent's sandbox, and the last file in your storage. Each copy changes on its own, and a save writes one copy over the others. With the server there is one copy, and every writer sends edits to it, not files.

| Problem with copies | With the document server |
| - | - |
| The last save wins, and the other writers' edits are lost | Edits merge. No writer saves a whole file |
| The agent edits the last saved file, without the user's unsaved typing | The agent's code runs on the current copy |
| The user must wait for the agent, or the agent's result is stale | The user types while the agent works. Its edit is rebased past the typing |
| Each service that reads the document must ask a tab for its copy | Every service reads the current `.docx` from the server |
| A closed tab loses its unsaved edits | Each edit is stored when the server accepts it. A tab that reconnects gets the edits it missed |

## When you need it

* More than one tab or user edits the same document.
* An agent edits a document while a person has it open.
* A background job reads or exports the document that a user is editing.

If one person edits one file at a time, you do not need the server. The [editor](https://docs.reducto.ai/office/docx/ui/editor.md) and the [JS API](https://docs.reducto.ai/office/docx/js-api/concepts.md) work on their own.
