> ## Documentation Index > Fetch the complete documentation index at: https://docs.tryrollup.com/llms.txt > Use this file to discover all available pages before exploring further. # SearchResults > What Rollup's on-device search returns. ```swift theme={null} public struct SearchResults: Codable, Sendable ``` What `Rollup.shared.search(_:)` returns: every result in order with what the index holds about it, the query as search read it, and how long it took. The search as asked. The results, most relevant first. How search read the query. Seconds for the whole search. Seconds for its parts, by name, such as `locationResolution`, `filtering`, `embedding` and `text`. What the index covers: counts, dates and media. ## SearchPlan ```swift theme={null} public struct SearchPlan: Codable, Sendable ``` `semantic_search`, `location_search`, `date_search`, `capture_search` or `holiday_search`. `content`, `place`, `content_and_place`, `place_type`, `holiday` or `metadata`. What was left to match images and text against once dates, media and places were read. `most_relevant` or `newest_first`. The whole plan: the date, capture, location and place filters search applied, with its policies and counts. ## SearchResult ```swift theme={null} public struct SearchResult: Identifiable, Codable, Sendable ``` One asset and what the index holds about it. The asset's PhotoKit local identifier. When it was taken. The time zone where it was taken. `.photo`, `.video` or `.screenshot`. Whether the person marked it as a favorite. Where it was taken, as the on-device geography placed it. What it shows. The venue its visit resolved to. Empty when `places` is off or no venue was resolved. Cuisines of its visit, on food photos. The text search matches against: sign text, text read from the preview, or a screenshot's text. What was read from a screenshot. The relevance model's probability, for image matches. The words of the asset's text that hold the query, for text matches. Search's score for this result. The retrieval score before ranking. How closely its place matched a place in the query. Why it matched, such as `image`, `text` or `place`. The named places the result matched. ### Location The neighborhood, city and country that are known, joined with commas. ### Label The label's concept ID. The label, such as `sushi` or `dog`. A probability: Apple's confidence, the dish model's, or the learned selector's. The image model's cosine similarity, for labels without a probability. ### Place The venue's name. The venue's category. How confident the place resolution is. ### Screenshot The kind of app the screenshot was taken in. What the screenshot is about. What the person was doing. Each field's top probability, by field: `app_kind`, `topic` and `action`. Kinds of sensitive content the screenshot holds. A date it shows, as `YYYY-MM-DD` or `--MM-DD`, with `THH:MM` when a time was shown. How the shown date relates to when the screenshot was taken. Entities Apple's on-device model read, each with an `id`, a `type` and a `value`. Whether entities were read, and if not, why. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.