Search

How search works

Gearbox Relay uses two complementary search systems. Full-text search (SQLite FTS5) indexes all capture text, summaries, and tags for keyword matching. Semantic search (LanceDB) stores embedding vectors that let you find captures by meaning.

Search interface

Desktop

Press Ctrl+K (or Cmd+K on macOS) to open the search palette from anywhere in the app. Start typing to see results immediately.

Mobile

Tap the search bar at the top of the screen. Results appear as you type.

Search operators

You can use several operators to refine your searches.

Keyword search finds captures containing specific words. Phrase search (using quotes) finds exact phrases. Prefix matching (using asterisks) finds words starting with a prefix. Tag filtering (using tag:) finds captures with specific tags.

Semantic search

Semantic search finds captures by meaning rather than exact words. Open the search palette, type a natural language query, and results are ranked by semantic similarity to your query. This is useful when you remember the gist but not the exact words.

Filters

Filter search results by date range, tags, capture source (clipboard, extension, share, manual paste), or Stream membership.

Search tips

Be specific when searching. Use semantic search for exploration when you are not sure what you are looking for. Combine filters for precise results. Tag names are indexed, so searching tag:project-x finds all captures with that tag.

Performance

Full-text search responds in under 50ms. Semantic search adds 10-30ms. Libraries with thousands of captures maintain sub-100ms search latency.

Limitations

Captures are indexed for full-text and semantic search, but some metadata (exact capture time, source app) is not keyword-searchable. Deleted captures are removed from the index immediately.