cache /api/search results for 3 minutes to skip repeat yt-dlp spawns

server/server.js

@@ -353,11 +353,25 @@ app.get('/api/version', (c) =>
   )
 );
 
+// Same query repeats a lot (users re-searching, several devices, the
+// landing-page chips) and yt-dlp's own search takes several seconds — cache
+// the combined (library + YouTube) result for a short TTL. Same shape as
+// resolveStreams()' streamCache below, just keyed by normalized query text
+// instead of videoId. Only a full success is cached, so a transient yt-dlp
+// failure still gets retried on the next request.
+const SEARCH_CACHE_MAX = 100;
+const SEARCH_CACHE_TTL_MS = 3 * 60_000;
+const searchCache = new Map(); // lowercased q -> { results, expiresAt }
+
 // GET /api/search?q=<query>
 app.get('/api/search', async (c) => {
   const q = (c.req.query('q') || '').trim();
   if (!q) return c.json({ ok: false, error: 'empty query' }, 400);
 
+  const cacheKey = q.toLowerCase();
+  const cached = searchCache.get(cacheKey);
+  if (cached && Date.now() < cached.expiresAt) return c.json({ ok: true, results: cached.results });
+
   // This server's own library first — and it still answers when YouTube
   // (yt-dlp) is unreachable or rate-limited.
   let mine = [];
@@ -368,7 +382,10 @@ app.get('/api/search', async (c) => {
       '--dump-json', '--flat-playlist',
       '--no-warnings', '--ignore-errors',
     ]);
-    return c.json({ ok: true, results: [...mine, ...parseCards(out)] });
+    const results = [...mine, ...parseCards(out)];
+    if (searchCache.size >= SEARCH_CACHE_MAX) searchCache.delete(searchCache.keys().next().value);
+    searchCache.set(cacheKey, { results, expiresAt: Date.now() + SEARCH_CACHE_TTL_MS });
+    return c.json({ ok: true, results });
   } catch (err) {
     if (mine.length) return c.json({ ok: true, results: mine, youtubeError: err.message });
     return c.json({ ok: false, error: err.message }, 500);