AI Agent Connected to Google Drive: What Is Documented

An AI agent connected to Google Drive knowledge base content keeps a synced copy of the Drive files a team selects and answers employees from it. Serval, Harmony, Atomicwork and Ravenna each document such a connection, and their pages differ on three points. What is indexed: Google Docs, Sheets and Slides plus uploaded PDF and Office files at Serval, Harmony and Atomicwork, and only shared drives at Harmony, whose page leaves out files held only in a personal My Drive. Who gets an answer: Serval, Harmony and Atomicwork say Drive sharing decides it, while Ravenna’s page scopes knowledge by folder, channel and agent. How fresh it is: Atomicwork states a sync every 24 hours, Ravenna every 24 hours by default and Harmony a schedule without an interval, while Serval’s pages give both a four-hour sync for connected sources and continuous indexing for Drive. Sharing changes follow a cadence of their own, which the pages state differently.

What Drive content does a connected agent read?

Each row restates what that vendor’s page says, not what the product may do beyond it.

Agent	What its page says it reads from Drive	Refresh
Serval	Lists Google Drive among nine sources, covering “Docs, Sheets, Slides, and uploaded PDF, Word, and Excel files”. Content is synced read-only, from the folders, spaces or pages chosen in Serval	Connected-sources page: “connected sources automatically sync every four hours”; Google Workspace page: “Drive content is indexed continuously on connections granted a Drive scope”. A manual sync is available
Harmony	Google Docs, Sheets and Slides plus uploaded files, taken from the shared drives an administrator selects, through a read-only connection. Nothing is indexed until someone chooses the documents	“on a recurring schedule”, no interval given
Atomicwork	Docs, slides, sheets and PDFs; workspace admins add Drive folders as Topics, and all files and subfolders in them are indexed	Every 24 hours, or by using the “sync now” option
Ravenna	Google Drive is one of eight sources that feed knowledge folders, scoped to agents and channels	Knowledge folders refresh every 24 hours by default, with a manual sync at any time

The table shows two ways of choosing content. Harmony and Serval start from a selection of drives or items, and Harmony then has someone choose which documents to index. Atomicwork indexes whole folders added as Topics. Ravenna imports documents into folders that are then scoped to agents and channels.

What do these pages say the agent will not read or may miss?

Limits belong to the product that states them, and none of the following is a property of Drive as a category.

• Harmony reads only the shared drives that the connected Google account is a member of. Files that live only in a personal My Drive are not synced. Google Forms, Google Sites, shortcuts, videos, images and trashed files are not synced, and files larger than 50 MB appear in the catalog without their content being indexed.
• Serval shows a “Some content isn’t searchable yet” warning when a sync could not reach part of a source, often after a rate limit, and a later sync picks up the rest.
• Ravenna marks a document that disappears from the source with an error and does not delete it, and its sync fails until the integration is reconnected if credentials expire.
• Atomicwork’s page lists supported file types (DOCX, PDF, PPTX, CSV, XLSX and the Google formats) and states that renamed files are updated and deleted files removed during the 24-hour cycle.

Does the agent apply Drive’s sharing settings to answers?

Among Drive’s sharing settings, two matter here. Google’s help page defines “Anyone with the link” as “Anyone who has the link can use your file, without signing in to their Google Account” and “Restricted” as “Only people with access can open the file.” Google’s API documentation adds that “All child files and folders automatically inherit permissions from their parent folder”, and that an inherited permission cannot be reduced on a child item: the change is made on the originating parent, or the folder uses the limited access setting.

What the four agents say they do with that:

• Serval states that it respects document permissions from the source app “when the integration supports propagating access controls”, naming Google Drive. It re-reads every synced file’s sharing settings at least once a day, and a manual sync re-reads them at once. If it can no longer read a Drive file’s sharing settings, it hides that file from everyone until it can again.
• Harmony records, per file, who the file is shared with, including sharing inherited from the drive and its folders, and filters answers at query time. People are matched by email address, and Google Groups are matched by name against the groups the identity provider sends. A group it does not know shows as “Unsynced groups”. “Anyone with the link” maps to everyone in the tenant.
• Atomicwork says Google Drive remains the source of truth for file permissions and that users can only access what they can see in Drive.
• Ravenna’s knowledge page does not describe Drive sharing. It describes access in terms of knowledge folders: an agent reaches a folder only if both the channel and the agent are allowed to.

Of the pages read here, only Harmony’s says what happens to a file open to anyone with the link.

How soon does an edit or a sharing change reach the answers?

Harmony, Atomicwork and Ravenna describe a scheduled sync for content, so on their pages an edit made in Drive can become available to the agent after the next successful run, with no promise that the next answer uses it. Serval’s two pages differ: its connected-sources page says “connected sources automatically sync every four hours”, and its Google Workspace page says “Drive content is indexed continuously on connections granted a Drive scope”. The two pages do not give a single cadence for Drive.

• 01

  Refresh intervals differ

  Serval states four hours for connected sources on one page and continuous indexing for Drive on another. Atomicwork states twenty-four hours, and Ravenna twenty-four by default for its knowledge folders. Harmony says a recurring schedule, without an interval.

  Vendor documentation, October 2026
• 02

  A link is not a person

  Google defines "Anyone with the link" as open without sign-in. Harmony's page maps it to everyone in the tenant. The other pages read do not say how they treat it.

  Google Drive Help; Harmony docs

Sharing changes have cadences of their own, which the pages state differently. Serval re-reads Drive sharing at least once a day, and a manual sync re-reads it at once. Atomicwork’s page says both that permissions carry over “whenever changes occur in Drive” and that permission updates sync every 24 hours; its “sync now” option is described for content. Harmony says sharing changes are picked up on the next sync, without an interval. Until that refresh runs, a permission change may not yet apply to answers, and how long that lasts depends on where the change falls in the cycle and on whether someone triggers a manual sync. A team that removes access to a sensitive file should run the manual sync where one is documented rather than wait.

Is a Google Workspace connection the same as a Drive connection?

It depends on the product. Harmony’s page shows two connections: Drive content is authorized through a Google account that is a member of the shared drives, with a read-only scope covering Drive files, and the page suggests a dedicated service user, a Google Workspace account added to those drives as a Viewer and used to sign in to the connection. That is a service account in the general sense, not a Google Cloud service account. Group names come from the identity provider that Harmony is connected to; when that provider is Google Workspace, the names match as long as the group is included in the groups sent to Harmony.

Serval’s Google Workspace page describes a single integration that syncs users and groups and also indexes Drive for knowledge search. On the identity side, the sync covers group members too and runs on a recurring schedule, Serval workflows can add or remove users in Google Groups, and “Sign in with Google” is enough for that user and group management. Drive indexing needs a Drive scope, so it works with a Google Cloud service account with domain-wide delegation or with the team’s own OAuth app, and not with “Sign in with Google”. Serval’s connected-sources page points to that same guide for its Drive step.

So a product that integrates with Google Workspace for users and groups has not, by that fact alone, been described as reading documents, and in Serval’s case the same integration is where Drive indexing is set up. What an identity connection carries, and how the Okta and Entra ID connectors compare, is covered on the identity providers page.

What should a team check before relying on a Drive source?

Start from what each page states, not from what a page omits: a page that does not list a source says what it lists, not that the source is impossible.

1. Ask the vendor, in writing, when its pages are silent or disagree, whether a Drive source exists, on which plan, and with which permission behavior, before moving any document. Siit is a case in point. In October 2026 its Quickstart lists, among recommended next steps, “Add Knowledge sources (Notion, Confluence, Google Drive) with RBAC retrieval and redaction”, while its knowledge base page says Siit “integrates natively with three knowledge platforms”, Notion, Confluence and Slab, and does not name Google Drive. The two pages do not establish a single position on Drive.
2. Test on a small set of the documents employees ask about most, with real questions and with accounts that have different Drive access, before widening the scope.
3. Treat copying as a workaround for one configuration. Copying Drive documents into a source the agent lists leaves two copies, and someone has to decide which one is canonical. Which documents to give an agent is the subject of the guide on training an agent on internal documentation, and how a wiki-based source behaves is set out on the knowledge sources page.
4. Check the documented fallbacks of the agent you test. Serval, for instance, points to custom ingestion workflows for a system that has an API and is not listed. That is one product’s route, and it is not a feature of the others.

Which vendors document a Google Drive knowledge connection?

Listed here: vendors whose public documentation, read in October 2026, names Google Drive as a knowledge source and states what is synced, how access to the synced content is governed, and when it refreshes. “When” is read as a frequency or as a trigger.

• Atlassian: its page on how Teamwork Graph connector permissions are kept in sync describes an admin-managed connector to a third-party app such as Google Drive. It states that Rovo syncs when content is created or removed in that app, that access restrictions are kept so that restricted content can only be seen and used in Rovo by people who already have access to it in the source app, and that the content is used in Rovo Search, Chat and Agents; it gives no interval.
• Harmony: its Google Drive page states that it syncs Docs, Sheets, Slides and uploaded files from selected shared drives on a recurring schedule and filters answers by each file’s Drive sharing.
• Ravenna: its knowledge overview lists Google Drive among the sources of knowledge folders, which sync every 24 hours by default and are scoped to channels and agents.
• Serval: its connected-sources page lists Google Drive among nine sources, states that connected sources sync every four hours and that Drive sharing is re-read at least once a day; its Google Workspace page says Drive content is indexed continuously once a Drive scope is granted.