Duplicate Radar
Duplicate Radar spots documents that cover the same ground, so overlapping pages get merged or linked before they drift apart.
Meldoc compares document content and groups likely duplicates into clusters. You’ll see them in two places: a badge on the document itself, and the Duplicates tab on your workspace Overview.
Note: Duplicate Radar is available on Pro and above, and detection runs on a schedule — new edits are picked up within a few minutes (up to about 15 while the workspace keeps being edited), not instantly, and an open document refreshes its matches when the scan finishes. Documents are compared across all projects of the workspace; in a workspace of more than 5000 documents, only within each project. Near-empty documents (under a couple of paragraphs) are left out, so template scaffolds don’t cluster with each other.
Review duplicates on a document
When a document has likely duplicates, a N likely duplicates chip appears on the page, along with a Possible duplicates section in the right sidebar.
- Select the chip to open the Possible duplicates popover.
- Review each match — a match from another project is tagged cross-project.
- Select Open to compare the other document, or Dismiss to hide the match.
Keep in mind: The document shows the same clusters as the Duplicates tab, and a dismissal is shared: dismissing a match on the document, or its cluster on the tab, hides it for everyone in the workspace. Turn on Dismissed only in the sidebar to find a match and Restore it.
Browse duplicate clusters
- On your workspace Overview, select the Duplicates tab. A number on the tab counts the clusters that still need a look — anything already dismissed, or labelled as nothing to do, is left out of it.
- Review the cluster cards — each shows how similar the documents are, plus tags like cross-project and already linked.
- Use Show already linked to widen the list, Dismissed only to review what you have already dismissed, and Load more to page through.
The tab shows when the last scan ran. Select Dismiss on a cluster you’ve reviewed. Dismissed only replaces the list with those clusters — it does not add them to it — so you can find one and Restore it.
Why documents came out similar
Similar is not the same as redundant. A cluster can carry a label saying what it actually is, and two of the five labels are reasons to leave the pages alone:
| Label | What it means | What to do |
|---|---|---|
| Same subject, twice | The pages cover the same ground, even if the wording differs. | Merge them into one. |
| Shared block | Different subjects that quote the same block — a warning, a policy, an inherited preamble. | Extract the shared part, if anything. Don’t merge the pages. |
| Parallel structure | The same kind of page repeated under different parents, like one Technologies page per service. | Nothing. This is how the tree is meant to look. |
| Not really similar | The match looks like a false positive. | Nothing. |
| Could not tell | There wasn’t enough to go on — usually pages that are mostly tables or UI labels. | Nothing, unless you disagree. |
Labelled clusters sort by what the label says: Same subject, twice rises to the top, and the two “leave it alone” labels sink and arrive collapsed. Collapsed, not hidden — select Show N documents to open one.
When the answer is certain, the cluster goes away entirely. A Parallel structure or Not really similar cluster that Meldoc is confident about is removed from the tab, from the document page, and from what agents see over MCP — the pages stop pointing at each other altogether. This is what makes a set of a dozen pages built from one template stop flagging each other on every single page.
Two things bound it. Only those two labels can do it, so nothing that suggests an action is ever hidden: Same subject, twice and Shared block always stay visible, and so does Could not tell. And Meldoc keeps re-checking a cluster it has hidden, so if the pages change and the answer changes with them, the cluster comes back on its own. If you believe something was hidden wrongly and it does not come back, ask support — an administrator can restore a pair permanently.
A cluster with no label is normal, not an error. It may be new, too short to be worth checking, or already dismissed. It sits between the labelled ones, counts towards the tab’s number, and behaves exactly as it did before labels existed.
Select Recheck on a cluster to look at it again straight away. Labels are otherwise refreshed in the background, at most once a day per cluster.
Note: Labels need an AI provider. Without one — or on a workspace still on its first scan — clusters simply arrive unlabelled, and everything else on the tab works as usual.
What’s next?
Magic Links (Wiki Links) — Link related documents instead of duplicating them.
Search & Ask AI — Find content across all your projects.