docs: apply similarity threshold to English Ask AI retrieval

English vector search returned the nearest chunks with no minimum score, so
off-topic queries got grounded on weak matches. Require cosine similarity >= 0.6
(mirrors the site search route) so uncovered questions return nothing.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: ouiliame

apps/docs/app/api/chat/route.ts

@@ -17,6 +17,9 @@ const SEARCH_LIMIT = 6
 /** Candidates pulled before locale filtering, so a locale still yields SEARCH_LIMIT results. */
 const SEARCH_CANDIDATES = SEARCH_LIMIT * 4
 
+/** Minimum cosine similarity for an English vector match (mirrors the site search route). */
+const SIMILARITY_THRESHOLD = 0.6
+
 /** Locales the docs are published in (mirrors the site search route). */
 const KNOWN_LOCALES = ['en', 'es', 'fr', 'de', 'ja', 'zh']
 const DEFAULT_LOCALE = 'en'
@@ -207,7 +210,9 @@ async function searchDocs(query: string, locale: string) {
     rows = await db
       .select(SEARCH_COLUMNS)
       .from(docsEmbeddings)
-      .where(localeFilter(locale))
+      .where(
+        sql`1 - (${docsEmbeddings.embedding} <=> ${vectorLiteral}::vector) >= ${SIMILARITY_THRESHOLD} and ${localeFilter(locale)}`
+      )
       .orderBy(sql`${docsEmbeddings.embedding} <=> ${vectorLiteral}::vector`)
       .limit(SEARCH_CANDIDATES)
   } else {