diff --git a/server/db.js b/server/db.js
index 88637a4..2cfdbdb 100644
--- a/server/db.js
+++ b/server/db.js
@@ -112,6 +112,17 @@ export async function initDb() {
     CREATE INDEX IF NOT EXISTS idx_media_lru
       ON media_cache (status, last_access);
 
+    -- Persistent search cache: the YouTube part of a search (up to 200 cards,
+    -- JSON), keyed by the lowercased query. Trimmed LRU by count and bytes.
+    CREATE TABLE IF NOT EXISTS search_cache (
+      q           TEXT    PRIMARY KEY,
+      results     TEXT    NOT NULL,
+      size        INTEGER NOT NULL DEFAULT 0,
+      fetched_at  INTEGER NOT NULL DEFAULT 0,
+      last_access INTEGER NOT NULL DEFAULT 0
+    );
+    CREATE INDEX IF NOT EXISTS idx_search_cache_lru ON search_cache (last_access);
+
     -- Shared per-video documents (kind = lyrics | chapters), visible to every
     -- user. The live copy is here; every save also lands in video_note_revs
     -- as a full snapshot, which is the server-side backup and undo history.
diff --git a/server/innertube.js b/server/innertube.js
index df1fcc0..929f1c3 100644
--- a/server/innertube.js
+++ b/server/innertube.js
@@ -8,10 +8,8 @@ export function lengthToSeconds(s) {
   return s.trim().split(':').map(Number).reduce((acc, n) => acc * 60 + n, 0);
 }
 
-export function parseSearch(json) {
-  const sections = json?.contents?.twoColumnSearchResultsRenderer?.primaryContents
-    ?.sectionListRenderer?.contents;
-  if (!Array.isArray(sections)) throw new Error('innertube: unexpected response shape');
+// Cards from a list of itemSectionRenderer-style sections.
+function cardsFromSections(sections) {
   const out = [];
   for (const s of sections) {
     for (const it of s?.itemSectionRenderer?.contents || []) {
@@ -34,6 +32,37 @@ export function parseSearch(json) {
   return out;
 }
 
+const tokenOf = (sections) => {
+  for (const s of sections || []) {
+    const t = s?.continuationItemRenderer?.continuationEndpoint?.continuationCommand?.token;
+    if (t) return t;
+  }
+  return '';
+};
+
+export function parseSearch(json) {
+  const sections = json?.contents?.twoColumnSearchResultsRenderer?.primaryContents
+    ?.sectionListRenderer?.contents;
+  if (!Array.isArray(sections)) throw new Error('innertube: unexpected response shape');
+  return cardsFromSections(sections);
+}
+
+// First page plus the token for the next one ('' when there is none).
+export function parseSearchPage(json) {
+  const sections = json?.contents?.twoColumnSearchResultsRenderer?.primaryContents
+    ?.sectionListRenderer?.contents;
+  if (!Array.isArray(sections)) throw new Error('innertube: unexpected response shape');
+  return { cards: cardsFromSections(sections), next: tokenOf(sections) };
+}
+
+// A continuation response: more cards plus the next token.
+export function parseContinuation(json) {
+  const items = (json?.onResponseReceivedCommands || [])
+    .flatMap((c) => c?.appendContinuationItemsAction?.continuationItems || []);
+  if (!Array.isArray(items) || !items.length) return { cards: [], next: '' };
+  return { cards: cardsFromSections(items), next: tokenOf(items) };
+}
+
 export async function search(q, { fetchImpl = fetch, timeoutMs = 6000 } = {}) {
   const res = await fetchImpl('https://www.youtube.com/youtubei/v1/search?prettyPrint=false', {
     method: 'POST',
@@ -44,3 +73,34 @@ export async function search(q, { fetchImpl = fetch, timeoutMs = 6000 } = {}) {
   if (!res.ok) throw new Error(`innertube: HTTP ${res.status}`);
   return parseSearch(await res.json());
 }
+
+// Up to `limit` results (default 200) by following continuation tokens. The pages
+// are a chain, so they are fetched one after another; every page is optional —
+// a failure part-way returns what was gathered. `onPage(cards)` lets a caller
+// save progress.
+export async function searchDeep(q, { limit = 200, fetchImpl = fetch, timeoutMs = 6000, maxPages = 14, onPage } = {}) {
+  const post = async (body) => {
+    const res = await fetchImpl('https://www.youtube.com/youtubei/v1/search?prettyPrint=false', {
+      method: 'POST', headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ context: { client: CLIENT }, ...body }), signal: AbortSignal.timeout(timeoutMs),
+    });
+    if (!res.ok) throw new Error(`innertube: HTTP ${res.status}`);
+    return res.json();
+  };
+  const first = parseSearchPage(await post({ query: q }));
+  const seen = new Set();
+  const all = [];
+  const add = (cards) => { for (const c of cards) if (!seen.has(c.id)) { seen.add(c.id); all.push(c); } };
+  add(first.cards);
+  if (onPage) onPage(all.slice());
+  let next = first.next;
+  for (let i = 0; i < maxPages && next && all.length < limit; i++) {
+    let page;
+    try { page = parseContinuation(await post({ continuation: next })); } catch { break; }
+    if (!page.cards.length) break;
+    add(page.cards);
+    if (onPage) onPage(all.slice());
+    next = page.next;
+  }
+  return all.slice(0, limit);
+}
diff --git a/server/innertube.test.js b/server/innertube.test.js
index 1f383a1..62dc2ce 100644
--- a/server/innertube.test.js
+++ b/server/innertube.test.js
@@ -1,6 +1,6 @@
 import { test, expect } from 'bun:test';
 import { readFileSync } from 'node:fs';
-import { parseSearch, lengthToSeconds, search } from './innertube.js';
+import { parseSearch, lengthToSeconds, search, searchDeep, parseContinuation } from './innertube.js';
 const fixture = JSON.parse(readFileSync(new URL('./fixtures/innertube-search.json', import.meta.url)));
 
 test('parses video renderers into slim cards', () => {
@@ -27,3 +27,23 @@ test('search() throws on HTTP error', async () => {
   const fetchImpl = async () => new Response('no', { status: 429 });
   await expect(search('x', { fetchImpl })).rejects.toThrow('429');
 });
+
+const card = (id) => ({ videoRenderer: { videoId: id, title: { runs: [{ text: 't' + id }] }, ownerText: { runs: [{ text: 'c' }] } } });
+test('searchDeep follows continuation tokens, de-duplicates and stops at the limit', async () => {
+  const calls = [];
+  const fetchImpl = async (_u, init) => {
+    const body = JSON.parse(init.body);
+    calls.push(body.continuation || 'first');
+    const mk = (ids, token) => [{ itemSectionRenderer: { contents: ids.map(card) } }, ...(token ? [{ continuationItemRenderer: { continuationEndpoint: { continuationCommand: { token } } } }] : [])];
+    if (!body.continuation) return Response.json({ contents: { twoColumnSearchResultsRenderer: { primaryContents: { sectionListRenderer: { contents: mk(['a', 'b', 'c'], 'T1') } } } } });
+    if (body.continuation === 'T1') return Response.json({ onResponseReceivedCommands: [{ appendContinuationItemsAction: { continuationItems: mk(['c', 'd', 'e'], 'T2') } }] });
+    return Response.json({ onResponseReceivedCommands: [{ appendContinuationItemsAction: { continuationItems: mk(['f', 'g'], '') } }] });
+  };
+  const all = await searchDeep('x', { fetchImpl, limit: 200 });
+  expect(all.map((c) => c.id)).toEqual(['a', 'b', 'c', 'd', 'e', 'f', 'g']);
+  expect(calls).toEqual(['first', 'T1', 'T2']);
+  expect((await searchDeep('x', { fetchImpl, limit: 4 })).length).toBe(4);
+});
+test('parseContinuation of an empty response is empty', () => {
+  expect(parseContinuation({})).toEqual({ cards: [], next: '' });
+});
diff --git a/server/search-cache.js b/server/search-cache.js
new file mode 100644
index 0000000..7fc5472
--- /dev/null
+++ b/server/search-cache.js
@@ -0,0 +1,60 @@
+// Persistent search cache (libsql). Holds the YouTube cards of a search so a
+// repeat query — from any device, or after a restart — never waits on YouTube.
+// Capped by row count (default 250,000 queries) AND bytes (default 2 GiB), LRU.
+import { db } from './db.js';
+
+export const MAX_ROWS = Number(process.env.SEARCH_CACHE_MAX_ROWS) || 250_000;
+export const MAX_BYTES = Number(process.env.SEARCH_CACHE_MAX_BYTES) || 2 * 1024 ** 3;
+export const PER_QUERY = 200;
+
+let writes = 0;
+
+export async function get(q) {
+  const key = q.toLowerCase();
+  const r = await db.execute({ sql: 'SELECT results, fetched_at FROM search_cache WHERE q = ?', args: [key] });
+  const row = r.rows[0];
+  if (!row) return null;
+  let results;
+  try { results = JSON.parse(row.results); } catch { return null; }
+  // Touch at most once a minute per row — reads must stay cheap.
+  db.execute({ sql: 'UPDATE search_cache SET last_access = ? WHERE q = ? AND last_access < ?',
+    args: [Date.now(), key, Date.now() - 60_000] }).catch(() => {});
+  return { results, fetchedAt: Number(row.fetched_at) };
+}
+
+export async function put(q, results) {
+  const json = JSON.stringify(results.slice(0, PER_QUERY));
+  const now = Date.now();
+  await db.execute({
+    sql: `INSERT INTO search_cache (q, results, size, fetched_at, last_access) VALUES (?,?,?,?,?)
+          ON CONFLICT(q) DO UPDATE SET results=excluded.results, size=excluded.size,
+            fetched_at=excluded.fetched_at, last_access=excluded.last_access`,
+    args: [q.toLowerCase(), json, json.length, now, now],
+  });
+  if (++writes % 25 === 0) trim().catch(() => {});
+}
+
+// Drop least-recently-used rows until under both budgets.
+export async function trim({ maxRows = MAX_ROWS, maxBytes = MAX_BYTES } = {}) {
+  const t = (await db.execute('SELECT COUNT(*) AS n, COALESCE(SUM(size),0) AS b FROM search_cache')).rows[0];
+  let n = Number(t.n), b = Number(t.b);
+  if (n <= maxRows && b <= maxBytes) return 0;
+  let dropped = 0;
+  while (n > maxRows || b > maxBytes) {
+    const batch = (await db.execute('SELECT q, size FROM search_cache ORDER BY last_access ASC LIMIT 500')).rows;
+    if (!batch.length) break;
+    const drop = [];
+    for (const r of batch) {
+      if (n <= maxRows && b <= maxBytes) break;
+      drop.push(r.q); n -= 1; b -= Number(r.size);
+    }
+    await db.batch(drop.map((q) => ({ sql: 'DELETE FROM search_cache WHERE q = ?', args: [q] })));
+    dropped += drop.length;
+  }
+  return dropped;
+}
+
+export async function stats() {
+  const t = (await db.execute('SELECT COUNT(*) AS n, COALESCE(SUM(size),0) AS b FROM search_cache')).rows[0];
+  return { queries: Number(t.n), bytes: Number(t.b), maxQueries: MAX_ROWS, maxBytes: MAX_BYTES };
+}
diff --git a/server/search-cache.test.js b/server/search-cache.test.js
new file mode 100644
index 0000000..7bafb5b
--- /dev/null
+++ b/server/search-cache.test.js
@@ -0,0 +1,20 @@
+import { test, expect } from 'bun:test';
+import * as sc from './search-cache.js';
+import { initDb } from './db.js';
+
+await initDb();
+const cards = (n) => Array.from({ length: n }, (_, i) => ({ id: 'v' + i, title: 't' + i }));
+
+test('put/get round-trips, case-insensitive, capped at 200', async () => {
+  await sc.put('Hello World', cards(300));
+  const r = await sc.get('hello world');
+  expect(r.results.length).toBe(200);
+  expect(await sc.get('nope')).toBeNull();
+});
+
+test('trim evicts least-recently-used rows over the row cap', async () => {
+  for (let i = 0; i < 6; i++) await sc.put('q' + i, cards(3));
+  const dropped = await sc.trim({ maxRows: 3 });
+  expect(dropped).toBeGreaterThan(0);
+  expect((await sc.stats()).queries).toBe(3);
+});
diff --git a/server/server.js b/server/server.js
index 0d07ffe..a4dc4d2 100644
--- a/server/server.js
+++ b/server/server.js
@@ -56,6 +56,7 @@ import { registerIntakeRoutes } from './p2p-intake.js';
 import { createP2pHub, holdersPayload, createRehydrator } from './p2p-hub.js';
 import { sha256Range } from './hash.js';
 import * as innertube from './innertube.js';
+import * as searchCacheDb from './search-cache.js';
 import QRCode from 'qrcode';
 import { createYtdlpPool } from './ytdlp-pool.js';
 import { dirname, join as pathJoin } from 'node:path';
@@ -396,69 +397,78 @@ app.get('/api/version', (c) =>
 // 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 = 200;
+const SEARCH_CACHE_MAX = 200;                     // hot in-memory layer
 const SEARCH_CACHE_TTL_MS = 10 * 60_000;          // fresh: answered from memory
-const SEARCH_CACHE_STALE_MS = 6 * 3600_000;       // stale-while-revalidate window
-const searchCache = new Map(); // lowercased q -> { results, expiresAt }
-const inflightSearches = new Map(); // lowercased q -> Promise<{ results, youtubeError? }>
+const SEARCH_CACHE_STALE_MS = 15 * 24 * 3600_000; // persistent copy: stale-while-revalidate for 15 days
+const searchCache = new Map(); // lowercased q -> { yt, fetchedAt }  (YouTube cards only)
+const inflightSearches = new Map(); // lowercased q -> Promise<{ yt, youtubeError? }>
 
-// One search, shared by every caller asking the same thing at the same time.
-function fetchSearch(q) {
+// The YouTube side of one search (up to 200 cards), shared by every caller asking
+// the same thing at the same time, and written to memory + the persistent cache.
+function fetchYoutube(q) {
   const key = q.toLowerCase();
   if (inflightSearches.has(key)) return inflightSearches.get(key);
   const p = (async () => {
-    // This server's own library first — and it still answers when YouTube
-    // (yt-dlp) is unreachable or rate-limited.
-    let mine = [];
-    try { mine = (await notesDb.listUploads({ q, limit: 20 })).map(uploads.card); } catch { /* library optional */ }
-    try {
-      let yt = [];
-      if (process.env.SEARCH_INNERTUBE !== '0') {
-        try {
-          yt = await innertube.search(q);
-        } catch (e) {
-          console.warn(`[search] innertube failed, using yt-dlp: ${e.message}`);
-        }
+    let yt = [];
+    if (process.env.SEARCH_INNERTUBE !== '0') {
+      try {
+        yt = await innertube.searchDeep(q, { limit: searchCacheDb.PER_QUERY });
+      } catch (e) {
+        console.warn(`[search] innertube failed, using yt-dlp: ${e.message}`);
       }
-      if (!yt.length) {
-        const out = await runYtdlpResilient([
-          `ytsearch${SEARCH_LIMIT}:${q}`,
-          '--dump-json', '--flat-playlist',
-          '--no-warnings', '--ignore-errors',
-        ], { pooled: true });
-        yt = parseCards(out);
-      }
-      const results = [...mine, ...yt];
-      if (searchCache.size >= SEARCH_CACHE_MAX) searchCache.delete(searchCache.keys().next().value);
-      searchCache.set(key, { results, expiresAt: Date.now() + SEARCH_CACHE_TTL_MS });
-      return { results };
-    } catch (err) {
-      if (mine.length) return { results: mine, youtubeError: err.message };
-      throw err;
     }
+    if (!yt.length) {
+      const out = await runYtdlpResilient([
+        `ytsearch${SEARCH_LIMIT}:${q}`,
+        '--dump-json', '--flat-playlist',
+        '--no-warnings', '--ignore-errors',
+      ], { pooled: true });
+      yt = parseCards(out);
+    }
+    const fetchedAt = Date.now();
+    if (searchCache.size >= SEARCH_CACHE_MAX) searchCache.delete(searchCache.keys().next().value);
+    searchCache.set(key, { yt, fetchedAt });
+    if (yt.length) searchCacheDb.put(q, yt).catch((e) => console.warn(`[search] cache write: ${e.message}`));
+    return { yt, fetchedAt };
   })().finally(() => inflightSearches.delete(key));
   inflightSearches.set(key, p);
   return p;
 }
 
-// GET /api/search?q=<query>
-// Fresh cache → instant. Stale (up to 6 h) → answered instantly from the old
-// copy while a refresh runs in the background, so a repeat search never waits
-// on yt-dlp/InnerTube. Only a full success is cached, so a transient failure
-// still gets retried on the next request.
+const searchLibrary = async (q) => {
+  // This server's own library first — it still answers when YouTube is unreachable.
+  try { return (await notesDb.listUploads({ q, limit: 20 })).map(uploads.card); } catch { return []; }
+};
+
+// GET /api/search?q=<query>[&refresh=1]
+// Memory → persistent cache (fresh 10 min, then stale-while-revalidate up to 15
+// days: answered instantly, refreshed in the background) → YouTube. `refresh=1`
+// (the app's "Update search" button) skips every cache. Only a full success is
+// cached, so a transient failure gets retried on the next request.
 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 cached = searchCache.get(q.toLowerCase());
-  const age = cached ? Date.now() - (cached.expiresAt - SEARCH_CACHE_TTL_MS) : Infinity;
-  if (cached && Date.now() < cached.expiresAt) return c.json({ ok: true, results: cached.results });
-  if (cached && age < SEARCH_CACHE_STALE_MS) {
-    fetchSearch(q).catch(() => { /* keep serving the stale copy */ });
-    return c.json({ ok: true, results: cached.results, stale: true });
+  const key = q.toLowerCase();
+  const mine = await searchLibrary(q);
+  const answer = (yt, extra = {}) => c.json({ ok: true, results: [...mine, ...yt], ...extra });
+  if (c.req.query('refresh') !== '1') {
+    let hit = searchCache.get(key);
+    if (!hit) {
+      const row = await searchCacheDb.get(q).catch(() => null);
+      if (row) { hit = { yt: row.results, fetchedAt: row.fetchedAt }; searchCache.set(key, hit); }
+    }
+    const age = hit ? Date.now() - hit.fetchedAt : Infinity;
+    if (hit && age < SEARCH_CACHE_TTL_MS) return answer(hit.yt, { fetchedAt: hit.fetchedAt });
+    if (hit && age < SEARCH_CACHE_STALE_MS) {
+      fetchYoutube(q).catch(() => { /* keep serving the stale copy */ });
+      return answer(hit.yt, { stale: true, fetchedAt: hit.fetchedAt });
+    }
   }
   try {
-    return c.json({ ok: true, ...(await fetchSearch(q)) });
+    const { yt, fetchedAt } = await fetchYoutube(q);
+    return answer(yt, { fetchedAt });
   } catch (err) {
+    if (mine.length) return c.json({ ok: true, results: mine, youtubeError: err.message });
     return c.json({ ok: false, error: err.message }, 500);
   }
 });