Skip to main content
Search in Zark goes beyond filenames and folders. It finds files by meaning, content, and intent — powered by AI indexing that understands what’s inside your files, not just what they’re called.

​
How Search Works

When you search for files, Zark combines multiple signals:
SignalDescription
Semantic searchMatches by meaning — “financial reports” finds files about budgets, revenue, and accounting
Keyword searchExact and partial text matching across file names and contents
Content typeFilter by file format — images, videos, documents, data files
AI-generated tagsTags automatically created during file processing
MetadataFile size, upload date, format, folder location

​
Searching via Chat

The most natural way to search is through conversation: 
{
  "messages": [
    {"role": "user", "content": "Find all documents about the Q3 marketing campaign"}
  ],
  "workspace_id": "wks-xxxxx",
  "tools": ["file_manager"],
  "tool_choice": "auto"
}
More examples: 
Show me images from last week

Find files tagged "finance"

Which video mentions the product launch?

Search for spreadsheets with revenue data
Zark uses semantic understanding to match your query — you don’t need to know exact filenames.

​
AI Auto-Tagging

Tags are generated automatically when files are processed. No manual tagging is required.

​
What Gets Tagged

SourceTags generated from
ImagesObjects, scenes, subjects, colors, visual context
DocumentsTopics, themes, entities (people, companies, places), key terms
VideosScenes, actions, speakers, topics discussed, visual elements
AudioTopics, speakers, key moments, entities mentioned
Data filesColumn names, data patterns, detected categories
Tags are used internally to power search. You don’t manage them manually — they’re applied automatically and improve search accuracy. When you search for “sunset photos,” Zark matches files tagged with sunset, outdoor, sky, golden hour — even if the filename is IMG_4521.jpg.

​
Content Understanding

Search doesn’t just match tags — it understands content:
Content typeWhat becomes searchable
DocumentsFull text, headings, section structure, tables, key terms
ImagesVisual descriptions, detected text (OCR), objects, scenes
VideosTranscripts, visual content, key moments, speaker statements
AudioTranscripts, speaker identification, topics discussed
Data filesColumn names, data summaries, detected patterns
This means you can find a specific moment in a video by asking about what was said, or locate a document by describing its contents — not its filename.

​
Search at Scale

Search works efficiently across large file collections — thousands or millions of files. Results are ranked by relevance, not just recency. No manual organization is required for files to be discoverable.

​
Streaming Events

When searching via the API with streaming enabled:
EventDescription
file_search_resultsList of matching files with metadata, thumbnails, and relevance scores
ai_chunkAI-generated summary of search results
ai_completeFinal response with organized results