Google Drive Connector

Connect your Google Drive account to upload and process files directly as an external knowledge base. Supports Google Workspace files (Docs, Sheets, Slides), Office files, PDFs, text files, CSVs, images, and more. Authentication is handled via Google OAuth 2.0 with automatic token refresh.

Setup

GOOGLE_CLIENT_ID=your-google-client-id
GOOGLE_CLIENT_SECRET=your-google-client-secret
# Optional: lets members sync whole folders (see below)
GOOGLE_SERVICE_ACCOUNT_KEY=/path/to/service-account-key.json

Picked files or the whole Drive

GOOGLE_DRIVE_ACCESS sets what the sign-in asks Google for:

With file, members always pick in Google’s file picker, opened with GOOGLE_CLIENT_ID. DocsGPT’s own file browser would only list what was picked before. An Internal Workspace app needs no review in either mode, so readonly suits an install used only inside one organization.

A connection keeps the access it signed in with. After switching modes, existing connections work as before until the member reconnects.

Sync whole folders with a service account

With GOOGLE_DRIVE_ACCESS=file, Google gives the sign-in a picked folder but not the files inside it. A service account reads them instead.

How folders stay private:

• DocsGPT reads a folder only after the member’s own sign-in can open it. With drive.file, that is a folder they picked, so a folder ID alone is not enough.
• Inside it, items whose sharing is restricted apart from the folder (limited access) are left out, and so are shortcuts to anything the member cannot open themselves.
• The service account only ever reads. Anything shared with it can be synced by any member who can also open it, so share the folders you mean to sync, not whole drives you don’t.

A folder that is not shared with the service account is skipped and the rest of the source still syncs. Once it is shared, the next sync picks it up. Without GOOGLE_SERVICE_ACCOUNT_KEY, the picker offers files only.

Files, folders and shared drives

• Folders: a selected folder is ingested with everything inside it, the folders inside it included, and keeps its layout. A sync picks up files added to it later. With GOOGLE_DRIVE_ACCESS=file, this needs the service account.
• Shared drives: Google’s file picker has a Shared drives tab, and DocsGPT’s own picker (GOOGLE_DRIVE_ACCESS=readonly) lists shared drives you are a member of.
• Shortcuts: a shortcut is followed to the file or folder it points to, and saved under the shortcut’s name. Something reached twice, such as a shortcut to a folder that is already selected, is read once.
• Names: two files with the same name in one folder are kept apart as report.pdf and report_1.pdf. Google Docs, Sheets and Slides are converted to .docx, .xlsx and .pptx.
• Skipped files: types DocsGPT cannot read (Forms, Drawings, archives, video) are left out. So are files Google will not export, such as a Google Doc over Google’s 10 MB export limit. One skipped or failed file does not stop the rest of the folder.

Calls to Google are retried, waiting longer after each attempt, when Google rate-limits them or has a temporary error.

Sync

A scheduled sync, or Sync now, first compares each file’s name, location and modified time with the last ingest. When nothing changed, it keeps the existing index and downloads nothing. Otherwise it downloads and indexes the whole selection again. Changing the source’s chunking settings also counts as a change. If a file could not be downloaded last time, the next sync reprocesses the source in full. The exception is a file Google will never export, such as one over the size limit: it does not force a full sync until it is edited.

Troubleshooting

• Google Drive is not offered — GOOGLE_CLIENT_ID or GOOGLE_CLIENT_SECRET is missing from the backend .env, so the connector stays off. Admin > Connectors lists which one.
• Authentication failed — Check that the redirect URI matches exactly and equals CONNECTOR_REDIRECT_BASE_URI, with no query parameters. Ensure the Google Drive API is enabled.
• Sign-in fails with “not an allowed origin” — The frontend runs on an origin the sign-in may not return to. Add it to CONNECTOR_ALLOWED_ORIGINS in the backend .env.
• Permission denied — Verify the OAuth consent screen is configured and the user has access to the target files.
• A folder syncs no files — With GOOGLE_DRIVE_ACCESS=file, it is not shared with the service account. The worker log names it (“not shared with …”). Share it as a Viewer and sync again.
• Folders cannot be selected in the picker — GOOGLE_SERVICE_ACCOUNT_KEY is not set, or the key cannot be read; the API log says which.
• The picker does not open — Enable the Google Picker API in the project of GOOGLE_CLIENT_ID.
• Files not processing — Check backend logs and verify that backend environment variables are correctly set.