Skip to main content
Three kinds of “search”:

Filter

Narrow a known set by exact-match criteria. “Show me tasks assigned to me with status open.” Use indexed Convex queries.

Full-text

Find by keyword in text fields. “Tasks containing ‘Q2 report’.” Use Convex full-text search.

Semantic / vector

Find by meaning, not exact words. “Tasks about end-of-quarter reporting.” Use Convex vector search.

​
Decision tree

1

Are you matching exact values? (status = open, assignee = me)

Filter. Indexed where clauses.
2

Are you matching keywords in text? (titles, body, descriptions)

Full-text search.
3

Do you want results that are conceptually related, not just keyword-matched?

Vector search (with embeddings).

​
Combining

Many UIs combine all three:
  • A search bar (full-text on title + body)
  • Filter pills (filter by status, assignee, priority)
  • An “Ask in natural language” mode (vector search for conceptual queries)

​
Filter UI patterns

Pill filters

Click pills to toggle filters. Active filters visible as chips above the list.

Sidebar filters

Persistent filter panel. Best for data-heavy UIs (admin, analytics).

Inline command

A cmdk-style command bar. Power-user feel; takes practice.

URL query params

Persist filter state in the URL so it survives refresh and is shareable.

​
Tips

Start with filter, add search later. Most “search” requests are really filter requests in disguise.
Debounce search inputs. 200ms is plenty. Avoids hammering the backend on every keystroke.

Full-text search

Convex implementation.

Vector search

Semantic implementation.

Algolia

Hosted search alternative.
Last modified on April 18, 2026