Enterprise data controls

Two settings let a company narrow what GPT Workspace can reach and where its prompts may be processed. Both are in the app under Settings > Data controls (gpt.space/app/settings/data_controls), on every plan, and apply to the signed-in Google account at once.

The full list of processors, retention periods and security practices is on the Trust Center. For a signed data processing agreement, see the DPA template.

Whole Drive (the default) lets the agent work with any file your Google account can open.

Selected folders only restricts it to the top-level folders of My Drive and the shared drives you tick, including every subfolder below them. A file is allowed when one of the chosen folders is among its ancestors; a file anywhere else is treated as if it did not exist.

Where the restriction is enforced:

Surface	Behavior outside the allowed folders
The Drive agent in the chat	Search results are filtered, and a tool that is asked for a file outside the scope answers that it is not accessible.
@-mentioned Drive files in the chat	The file is not read, even when you paste its link or pick it from the picker.
The Google Drive MCP server (Claude, ChatGPT, Cursor and other clients)	Every tool checks the scope: get, search, tree view, copy, move, share, trash, revisions, metadata.
Creating a folder or a file	Only inside an allowed folder; the Drive root is not a valid parent in this mode.

Pick at least one folder before saving. Folders that are no longer at the top level of your Drive (moved, or a shared drive you left) stay in the list with their id so you can untick them.

The Google permission the app holds does not change: the restriction is enforced by GPT Workspace on every request, in addition to the Google permission. To revoke the permission itself, see How to connect Google account permissions.

What it does not cover:

• Google Docs, Sheets, Slides and Gmail tools that work on the current file or e-mail you opened yourself (the sidebar, the AI bars, the Gmail cards): you chose that file by opening it.
• Files you attach to a chat from your computer.

Every model request leaves GPT Workspace through OpenRouter, the gateway that routes it to the provider pinned for that model. Any provider (the default) uses the routing chosen for speed and quality for each model.

Zero-data-retention providers only adds OpenRouter’s zero-retention policy to every request (zdr: true with data_collection: deny): the request may only be served by endpoints that do not store prompts and do not train on them. The provider pins are then relaxed to a preference order, so a model whose usual endpoint keeps data can still be answered by one that does not.

Consequences to expect:

• Some models answer slower, because the fastest endpoint for them may be excluded.
• A request can fail when no zero-retention endpoint exists for the model it needs. Switch the setting back, or pick another model where the feature offers one.

What it covers: the chat and the agents, GPT formulas in Sheets, slide generation, image generation, web search (Perplexity runs through OpenRouter) and the MCP servers that call a model.

What it does not cover:

• Files attached to a chat or to a custom assistant. They are uploaded to an OpenAI vector store so the model can search inside them, and OpenAI must hold the file to do that. The store expires one day after it was last used; the files are deleted within a month. If a file may not leave Google, do not attach it: ask the Drive agent instead, which reads it in place.
• Stock image search for generated slides (Google Programmable Search receives the image query only).

The settings are per Google account. A Workspace administrator who needs them applied to every seat of a team subscription can write to contact@gpt.space with the domain and the folder ids; see Team subscription.

Related: Data and privacy, Delete your account and data, Questions before you subscribe.