Here's a problem every enterprise has and almost no one talks about.
The same document exists in fourteen places. There's the original in SharePoint. A copy someone emailed around. A version that was updated in Teams. One that was downloaded and re-uploaded six months ago. Three more attached to ServiceNow tickets. And the one that legal actually approved - which may or may not be any of the above.
When your AI retrieves "the policy on vendor security requirements," which one does it find? When a new hire asks the AI Assistant for the employee handbook, which version does it surface?
This is the canonical version problem, and it's one of the most consequential - and least discussed - issues in enterprise AI.
How vector databases make it worse
The standard answer - a vector database - doesn't solve this problem. It makes it worse.
To find the canonical version with a vector database, you'd copy every document into the store, embed all the versions, try to infer which is canonical based on timestamps, similarity, metadata, edit history, permissions, and user behavior, build workflows around that inference, and then keep everything synchronized. Forever.
That's a lot of plumbing for a problem users thought search should already solve.
SWIRL's approach: cluster, then ratify
SWIRL takes a different approach. Leave the data where it is. Query all sources simultaneously for the document. SWIRL then automatically clusters related results - email attachments next to OneDrive copies, SharePoint versions next to PDFs, drafts next to published versions - and surfaces the cluster with an AI-suggested canonical result.
But here's the important part: in most organizations, "canonical" is not just a technical decision. It's a business decision. Legal decides which contract version is approved. Marketing decides which pricing deck is current. Product decides which roadmap is authoritative.
So SWIRL lets organizations ratify those decisions directly in search, via a workflow called Pinned Results. An administrator or designated reviewer can pin a specific result as the canonical answer for a given query pattern. Everyone who searches for that document subsequently gets the pinned, authoritative version at the top.
The technical layer finds the candidates. The business layer ratifies the answer. And from that point forward, the AI serves the right version - not just any version - to everyone who asks.