.webp)
Avaratak Blog
Nobody Remembers Which Folder: Rovo Now Reads Google Drive and SharePoint
The project plan lives in Google Drive. The budget model is in SharePoint. The decision that reconciled the two is a Confluence page nobody linked to, and the work it produced is a Jira epic that three teams are quietly interpreting differently.
Nothing is missing. That is the frustrating part. Every answer your team needs already exists, written down, by someone who was paying attention at the time. The tax is not authorship. The tax is reassembly — and it gets paid again every time somebody prepares for a review.
Atlassian’s July 30 announcement goes after exactly that tax. Teamwork Graph connectors for Google Drive and Microsoft SharePoint now pull full document content — not just filenames and metadata — into the graph that powers Rovo Search, Rovo Chat, and Rovo Agents. Since the Teamwork Graph went public in May, it is the most consequential thing added to it — and a release where the setup instructions deserve more attention than the demo video.
Start with the outcome, not the folder
The behavioural shift is small and worth naming. Today, finding something means first remembering where it was saved — a retrieval problem you solve with memory before you solve it with a search box.
With the connectors active, the question changes shape. Instead of hunting for the onboarding plan, you ask Rovo to summarise the status of a project, flag inconsistent information across sources, and link back to whatever it drew from. Atlassian’s example prompts lean hard on this pattern, and it is the right instinct. We made a similar argument when Atlassian first started closing the context gap between AI and your actual business: the model is rarely the bottleneck. The context is.
What makes this more interesting than a federated search bar is that the document body comes along. Rovo can relate a Drive planning doc to the Confluence page recording the decision and the Jira work that followed, then cite all three. That is a different capability from the cross-company Ctrl+F we covered during Rovo Week, and a meaningful step past the keyword karaoke problem in Atlassian search.
It writes back, too
Once a connector is live, Rovo can also create a Google Doc or SharePoint page and comment on existing files. Atlassian is clear these actions respect the same permissions as search and require your direction. Worth piloting narrowly before anyone gets excited about agents filing documents unattended.
The number nobody puts on the slide
Here is the part I would want a client to see before the pilot, not after.
Third-party content pulled in through a Teamwork Graph connector counts as an indexed object, and your subscription includes an allowance that scales with seats and edition. Atlassian documents it plainly: 250 indexed objects per user on Jira Premium or Confluence Premium, 625 on Enterprise, 100 on Standard, with higher figures for Teamwork Collection. Allowances pool at the organisation level.
Run the arithmetic. Five hundred people on both Jira Premium and Confluence Premium gives you roughly 250,000 pooled indexed objects. Now consider what the Google Drive connector actually ingests: Docs, Sheets, Slides, PDFs under 10 MB, text files, CSVs, images, videos, and folders. A five-hundred-person company’s Drive does not contain 250,000 of those things. It contains several times that.
Credit where it is due: indexed objects do not consume Rovo credits, and the count appears in Admin Hub for visibility rather than billing. The documentation does not describe what happens at the ceiling. The practical conclusion is the same either way: “index everything and see what happens” is not a strategy. Scope deliberately, measure, expand. Atlassian’s own guidance says start narrow. The allowance table quietly says the same thing for a second reason.
Setup is not a toggle
The blog post says an organisation admin connects the source. True, and slightly compressed. What that actually involves:
- Google Drive requires domain-wide delegation in Google Workspace for Atlassian’s client ID and OAuth scopes. Without it, the connector only sees files owned by the single connecting account. Atlassian sensibly recommends a dedicated read-only admin account rather than routing calls through a Super Admin.
- SharePoint requires global admin rights, admin consent in Entra, and one easily-missed prerequisite: M365 site groups must allow everyone to view group membership. Without it, permissions for content tied to those groups will not sync correctly — the kind of detail that produces a confusing pilot two weeks later.
- Every individual still has to link their own account. The org connector handles indexing; the per-user link is what mirrors each person’s real access. Until someone connects, they may see nothing.
Expect a few hours before content appears, longer for large libraries. Encrypted SharePoint documents are not ingested at all. Videos index metadata only.
The governance controls are the good news
Both connectors let admins scope what enters the graph: site and drive allowlists or blocklists, personal-drive scoping by group, and date limits so only content modified after a chosen date is ingested.
SharePoint goes further with Microsoft Information Protection sensitivity label blocking, which works at both document and site level. Because labels travel with the file, protection follows content that moves between sites — something a site blocklist cannot do. Atlassian notes that already-indexed content carrying a newly blocked label stops appearing immediately and is purged on the next full scan.
Google Drive does not have label blocking yet. Atlassian lists it as coming soon. If your Drive governance depends on labels, that gap is your answer on sequencing: start with SharePoint.
Underneath all of it, Rovo respects source-system permissions. People see only what they could already open. Worth knowing the Drive nuance, though: a file surfaces in Rovo if its sharing is set to appear in search results, or if it is link-restricted and that link appears somewhere Rovo can read. Genuinely private files stay with their owner. This is the same governance-before-enthusiasm posture we applied to Trello’s new AI front door, and it holds up here.
The Avaratak Take
This is a strong release, and I would not talk a client out of it. Connecting document knowledge to the Teamwork Graph is the most direct way to make Rovo answers better, because it fixes the input rather than the model.
What I would talk them out of is treating it as a switch. Three things decide whether this lands:
- Pick one workflow. One recurring review that currently costs somebody an hour of collation. Instrument it, then judge the connector on that.
- Scope before you sync. Allowlist the sites and drives that matter. Use the date limit. Configure sensitivity label blocking on SharePoint before the first ingest, not after.
- Budget for the identity work. Domain-wide delegation and Entra admin consent are conversations with people who did not read the Atlassian blog and will reasonably want to know why.
The honest framing is this: your organisation already paid to create this knowledge. The connectors are a way to stop paying for it twice. Just do not let a five-minute admin flow write cheques your governance review has to cash.
If you want a second opinion on connector scoping before you turn one on, that is exactly the kind of question Avaratak is built for. We are an Atlassian Solution Partner staffed by senior-only consultants, and we would rather tell you to wait than sell you a rollout you are not ready for.