Full-text search

BM25-style search across text fields. Built into Convex, no external service.

For text search (titles, descriptions, comments) where keyword match matters more than semantic similarity, use Convex’s full-text search.

Setup

1. Add a search index

convex/schema.ts

tasks: defineTable({
  title:    v.string(),
  body:     v.string(),
  teamId:   v.id("teams"),
  status:   v.union(v.literal("open"), v.literal("done")),
})
  .searchIndex("search_text", {
    searchField:  "title",
    filterFields: ["teamId", "status"],
  })

2. Query

convex/tasks.ts

export const search = query({
  args: { query: v.string(), teamId: v.id("teams") },
  handler: async (ctx, { query, teamId }) => {
    return await ctx.db
      .query("tasks")
      .withSearchIndex("search_text", q =>
        q.search("title", query).eq("teamId", teamId)
      )
      .take(20);
  },
});

How it ranks

Convex uses a BM25-derived scoring algorithm — more matches in the search field rank higher; rarer words get more weight than common ones.

For semantic similarity (find similar items, not just keyword matches), use vector search.

Limits

• One search field per index. To search both title and body, either index a concatenated field or maintain two search indexes.
• Filter fields combined with search support up to 16 fields.

Tips

Search a concatenated field for “search across multiple columns.” Add a searchableText field that’s title + " " + body + " " + tags.join(" "), populate it on every write.

For autocomplete UIs, debounce the client. A keystroke-by-keystroke search is wasteful; 200ms debounce is plenty.

Related

Vector search

Semantic alternative.

Algolia

Hosted search with more features.

Schema

Search index syntax.