Fuzzy Search

Use this pattern for fuzzy search over a small data set. D1 returns a bounded candidate list, and Fuse.js ranks matches inside the Worker.

This approach accepts spelling errors without a new search service. It does not fit a large table or a high-traffic search endpoint.

1. Install Fuse.js

npm install fuse.js

2. Add the search function

Create src/lib/search-projects.ts. Replace project and its fields with your application table and fields.

import { desc } from 'drizzle-orm';
import Fuse from 'fuse.js';
import { project } from '@/config/schema';
import { db } from '@/db';

const MAX_CANDIDATES = 500;
const MAX_RESULTS = 20;

export async function searchProjects(rawQuery: string) {
  const query = rawQuery.trim().slice(0, 100);

  if (query.length < 2) {
    return [];
  }

  const candidates = await db
    .select({
      id: project.id,
      title: project.title,
      summary: project.summary,
    })
    .from(project)
    .orderBy(desc(project.updatedAt))
    .limit(MAX_CANDIDATES);

  const index = new Fuse(candidates, {
    keys: [
      { name: 'title', weight: 0.7 },
      { name: 'summary', weight: 0.3 },
    ],
    threshold: 0.35,
    ignoreLocation: true,
  });

  return index
    .search(query, { limit: MAX_RESULTS })
    .map((result) => result.item);
}

threshold controls the allowed difference. A smaller value requires a closer match.

3. Use the function from a route

Create src/pages/api/search.ts:

import type { APIRoute } from 'astro';
import { searchProjects } from '@/lib/search-projects';

export const GET: APIRoute = async ({ url }) => {
  const query = url.searchParams.get('q') ?? '';
  const results = await searchProjects(query);

  return Response.json({ results });
};

Protect private results before this route returns them. Add a rate limit before the database query for a public endpoint.

For live search, wait at least 250 ms after the last input before you send a request.

4. Know the limit

This query reads at most 500 rows for each search. When rows are large or searches are frequent, lower MAX_CANDIDATES.

When the table no longer fits this limit, use D1 FTS5. FTS5 can reduce the candidate set before Fuse.js ranks it.

When you need large indexes, language analysis, facets, or complex ranking, use a dedicated search service.

Read the Fuse.js guide for all search options.