StructorDocs
Local discovery Text + visual

Search by meaning

Semantic Search blends text meaning, visual meaning, and literal words. Ask for “receipts from April,” “beach photos,” or “contract drafts” instead of remembering exact filenames.

Structor · Semantic Search
Structor Semantic Search screen with folder and relevance controls
Search combines selected folder scope, optional file extensions, a relevance threshold, previews, and an explainable score breakdown.

Unlock the Search screen

Both prerequisites must be true:

1 of 2

A folder opted in

Open a folder's Settings → AI helpers and enable Semantic search. Only opted-in folders appear in Search scope.
2 of 2

Local assets ready

Open AI Model → Local search models and wait until MiniLM and SigLIP both report Ready.

How indexing works

When a folder opts in, Structor queues supported files for local analysis. Organization runs update paths in the index as files move.

StatusMeaningAction
IndexedSearchable nowNone
QueuedWaiting for processingLeave Structor running
ActiveProcessing nowWait for completion
FailedProcessing could not finishRead Last error, then reindex

Use Reindex from folder settings after large external changes, a replaced search model, or stale results.

  1. 01

    Write a natural query

    Describe content, subject, scene, or purpose. Meaningful phrases work better than comma-separated keywords.
  2. 02

    Choose folder scope

    Open Options and select one or more folders. Select all searches every folder with Semantic search enabled.
  3. 03

    Narrow file types

    Enter comma-separated extensions such as .pdf, .png, .txt. Leave the field empty for all supported files.
  4. 04

    Tune relevance

    Raise the threshold to hide weak matches. Lower it when a valid result may use different wording or visual context.
  5. 05

    Open or reveal

    Select a preview to open the result, or use Reveal in file manager. Video-shot results open the source video at the matched timestamp.

Understand results

Each result can include filename, path, extension, size, preview text, image/video thumbnail, timestamp, and several relevance signals:

Text

Semantic similarity between query meaning and extracted text.

Visual

Similarity between query meaning and image or video-frame content.

Lexical

Literal word or token overlap that strengthens exact wording matches.

Overall

Combined relevance used to rank and filter the result list.

Previews

  • Images display generated thumbnails.
  • Text files and PDFs can show extracted preview text.
  • Video matches represent indexed frames and include a timestamp label.
  • Unsupported preview formats still appear as normal file results.
Structor documentation