Amazon S3 Vectors gained metadata pre-filtering on September 30, changing how a search finds similar content inside a restricted document set. An enhanced index resolves the filter first, then searches the vectors that match it.
That helps when an agent needs one customer’s records from a much larger collection. The search runs over the eligible subset instead of drawing candidates from the full index while checking the filter along the way. AWS reports better recall for highly selective queries, meaning more relevant matching material is returned.
Existing indexes remain in Classic mode until explicitly updated to Enhanced. AWS says the change works in place, without re-ingesting vectors or changing existing queries, and adds no separate feature charge. Standard storage and request charges still apply.
The update also adds prefix matching for paths and other hierarchical identifiers. Queries support up to 100 filter constraints. A filter scopes the search; the application must still enforce who is allowed to access the underlying records rather than treating a user-supplied filter as authorization.