Skip to main content

Basic Search

The search engine is Octostar's interface for finding records across your data. It runs as a standalone application embedded inside Octostar. Under the hood, it is powered by OpenSearch, which indexes every record in the system for fast full-text and semantic retrieval.

Interface layout​

When you open the search view, you see:

  • A search bar at the top
  • A collapsible filter sidebar on the left
  • A results area filling the rest of the screen Type a query in the search bar and results appear with a short debounce delay.

Search modes​

The search engine supports four modes:

Text (default)​

Full-text search that breaks your query into terms and matches them across record fields using a layered relevance strategy:

  1. Exact matches on the record's display name (highest score)
  2. Phrase matches
  3. All-terms-present matches
  4. Broad search across all fields Wrap part of your query in quotes to search for an exact phrase.

Boolean​

Use explicit operators (AND, OR, NOT) for precise query construction.

Semantic​

Uses vector embeddings to find records conceptually similar to your query, even without exact word matches. The system converts your search text into a numerical embedding and runs a nearest-neighbor search against pre-indexed embeddings.

Auto​

An AI model reads your query and selects the most appropriate mode. A natural-language question might route to semantic search, while a keyword-heavy query routes to text mode. Auto mode shows you which mode it selected, and you can override it.

View modes​

Display results in three view modes:

  • Grid -- cards in a responsive masonry layout that adjusts the number of columns to your screen width
  • List -- one result per row with more detail per item
  • Table -- traditional columnar format with default columns for label, Workspace, last modified, created by, size, and notes. Add extra columns from the Concept's properties, resize them, or remove them. Column preferences are saved locally and persist between sessions. Switch between views at any time using the segmented control above the results.

Result rendering​

Each result is rendered using a template system that adapts to the record's Concept type. Templates control what information appears: display name, icon or thumbnail, Workspace badge, and Concept-specific properties. Search terms are highlighted in the results. Results are paginated at 20 items per page, with navigation controls at both the top and bottom of the results area.

Selecting results​

Select one or more results by clicking their checkboxes (in grid and list views) or row selectors (in table view). Hold Ctrl or Cmd and drag across results to draw a selection rectangle. When you have a selection, a toolbar appears showing the count of selected items and offering actions, including an Ask AI button that opens a chat session with your selected records as context. If the search was opened in a modal context (for example, when another part of Octostar asks you to pick records), clicking a result selects it directly rather than opening it, and a submit button sends your selection back to the calling view.

When the image detection API is available, a camera button appears in the search bar. Click it to open a full-screen modal where you upload an image. The system finds visually similar records using face detection and image similarity algorithms.

Saved searches​

Save a search by clicking the Save button. Octostar captures the current query (the full OpenSearch query that was last executed) and stores it as a Workspace item. An AI model generates a descriptive name for the saved search based on the query contents. Reopen saved searches later to re-execute the same query.