Autocomplete & Typeahead Feature Template

For entity search (people, projects, documents), start after 1 character to give immediate feedback. For general text search, 2 characters is a better default because single-character queries produce too many irrelevant suggestions. Test with your actual data to find the right threshold.

It depends on dataset size. If you have fewer than 10,000 entities, a client-side approach (download the index on page load, filter locally) gives zero-latency responses and eliminates network round trips. Above 10,000, a server-side index with a fast prefix lookup (trie, inverted index, or Elasticsearch completion suggester) is more practical. The [Search Relevance Template](/templates/search-relevance-template) covers server-side indexing patterns in detail.

Use fuzzy matching with a Levenshtein distance of 1 for queries under 5 characters and distance 2 for longer queries. Apply fuzzy matching only when exact and prefix matches return fewer than 3 results. This prevents fuzzy suggestions from overwhelming good exact matches.

150-250ms covers most use cases. Below 150ms, you fire too many requests. Above 300ms, users perceive lag. Adaptive debounce (shorter delay after slow typing, longer after fast typing) is an optimization worth testing once the basic implementation is stable.

If the user has active filters (e.g., "Status: Active"), autocomplete should respect those filters. Suggestions should only show results that match the current filter context. This prevents the frustrating experience of selecting a suggestion and getting zero results because it does not match the active filters. ---