diff --git a/sort_ym/lyrics.py b/sort_ym/lyrics.py
new file mode 100644
index 0000000..f440d7d
--- /dev/null
+++ b/sort_ym/lyrics.py
@@ -0,0 +1,87 @@
+from __future__ import annotations
+
+import json
+import time
+from pathlib import Path
+
+from yandex_music import Client
+from yandex_music.exceptions import NotFoundError
+
+from . import language
+from .ymclient import with_retries
+
+LYRICS_TEXT_CACHE_FILE = "lyrics_text.json"
+
+
+def ru_lyric_track_ids(tracks_cache: dict[str, dict], lang_cache: dict[str, str | None]) -> list[str]:
+    """Числовые id RU-треков, у которых вообще есть текст.
+
+    Гейт двойной: lyrics_available (иначе track_supplement заведомо вернёт запись без lyrics)
+    и RU - анализ INT-лирики вне скоупа фичи. Дедупликация нужна потому, что один и тот же
+    трек может быть лайкнут с разных альбомов: ключи tracks_cache при этом разные, а числовой
+    id (и, соответственно, ключ кэша текстов) - один.
+    """
+    ids: list[str] = []
+    seen: set[str] = set()
+    for t in tracks_cache.values():
+        if not t.get("lyrics_available"):
+            continue
+        tid = str(t["id"])
+        if tid in seen:
+            continue
+        lang = language.detect_language(t["title"], t["artists"], t["genre_raw"], lang_cache.get(tid))
+        if lang == "RU":
+            seen.add(tid)
+            ids.append(tid)
+    return ids
+
+
+def _atomic_write_json(path: Path, data: dict) -> None:
+    tmp = path.with_suffix(".tmp")
+    tmp.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+    tmp.replace(path)
+
+
+def load_lyrics_cache(cache_dir: Path) -> dict[str, str | None]:
+    cache_file = cache_dir / LYRICS_TEXT_CACHE_FILE
+    if cache_file.exists():
+        return json.loads(cache_file.read_text(encoding="utf-8"))
+    return {}
+
+
+def _fetch_one_lyrics(client: Client, track_id: str) -> str | None:
+    try:
+        supplement = with_retries(lambda: client.track_supplement(track_id))
+    except NotFoundError:
+        return None
+    if supplement is None or supplement.lyrics is None:
+        return None
+    return supplement.lyrics.full_lyrics
+
+
+def fetch_lyrics_text(
+    client: Client,
+    track_ids: list[str],
+    cache_dir: Path,
+    delay: float,
+) -> dict[str, str | None]:
+    cache_dir.mkdir(parents=True, exist_ok=True)
+    cache_file = cache_dir / LYRICS_TEXT_CACHE_FILE
+    cache = load_lyrics_cache(cache_dir)
+
+    missing = [tid for tid in track_ids if tid not in cache]
+    if not missing:
+        return cache
+
+    print(f"Запрос текста песни для {len(missing)} треков...")
+    for i, tid in enumerate(missing, 1):
+        cache[tid] = _fetch_one_lyrics(client, tid)
+        if i % 20 == 0:
+            _atomic_write_json(cache_file, cache)
+            print(f"  загружено {i}/{len(missing)}")
+        time.sleep(delay)
+
+    _atomic_write_json(cache_file, cache)
+    without_text = sum(1 for tid in track_ids if cache.get(tid) is None)
+    print(f"  готово: {len(track_ids)} треков, без текста {without_text}")
+    return cache
diff --git a/tests/test_lyrics.py b/tests/test_lyrics.py
new file mode 100644
index 0000000..48e67c9
--- /dev/null
+++ b/tests/test_lyrics.py
@@ -0,0 +1,119 @@
+import json
+from pathlib import Path
+
+import pytest
+from yandex_music.exceptions import NotFoundError
+
+from sort_ym import lyrics
+
+
+def track(id_, title="Песня", artists=("Артист",), genre_raw="rock", lyrics_available=True):
+    return {
+        "id": id_,
+        "album_id": 10,
+        "title": title,
+        "artists": list(artists),
+        "artist_ids": [7],
+        "genre_raw": genre_raw,
+        "lyrics_available": lyrics_available,
+    }
+
+
+def test_ru_lyric_track_ids_keeps_ru_and_drops_int():
+    tracks_cache = {
+        "100:10": track(100, title="Тоска", artists=["Гражданская Оборона"]),
+        "200:10": track(200, title="Come as You Are", artists=["Nirvana"]),
+    }
+
+    ids = lyrics.ru_lyric_track_ids(tracks_cache, lang_cache={})
+
+    assert ids == ["100"]
+
+
+def test_ru_lyric_track_ids_trusts_api_language_over_alphabet():
+    # Кириллическое название, но API говорит "en" - латиницы в тексте нет, а язык известен точно.
+    tracks_cache = {"100:10": track(100, title="Тоска", artists=["Артист"])}
+
+    ids = lyrics.ru_lyric_track_ids(tracks_cache, lang_cache={"100": "en"})
+
+    assert ids == []
+
+
+def test_ru_lyric_track_ids_skips_tracks_without_lyrics():
+    tracks_cache = {"100:10": track(100, title="Тоска", lyrics_available=False)}
+
+    assert lyrics.ru_lyric_track_ids(tracks_cache, lang_cache={}) == []
+
+
+def test_ru_lyric_track_ids_deduplicates_same_numeric_id():
+    # Один и тот же трек лайкнут с двух альбомов - числовой id общий, ключи кэша разные.
+    tracks_cache = {
+        "100:10": track(100, title="Тоска"),
+        "100:20": track(100, title="Тоска"),
+    }
+
+    assert lyrics.ru_lyric_track_ids(tracks_cache, lang_cache={}) == ["100"]
+
+
+class FakeLyrics:
+    def __init__(self, full_lyrics):
+        self.full_lyrics = full_lyrics
+
+
+class FakeSupplement:
+    def __init__(self, full_lyrics):
+        self.lyrics = FakeLyrics(full_lyrics) if full_lyrics is not None else None
+
+
+class FakeClient:
+    def __init__(self, texts):
+        self._texts = texts
+        self.requested: list[str] = []
+
+    def track_supplement(self, track_id):
+        self.requested.append(track_id)
+        if track_id not in self._texts:
+            raise NotFoundError("no such track")
+        return FakeSupplement(self._texts[track_id])
+
+
+def test_fetch_lyrics_text_caches_and_persists(tmp_path: Path):
+    client = FakeClient({"100": "Первая строка\nВторая строка"})
+
+    cache = lyrics.fetch_lyrics_text(client, ["100"], tmp_path, delay=0)
+
+    assert cache["100"] == "Первая строка\nВторая строка"
+    assert lyrics.load_lyrics_cache(tmp_path) == cache
+
+
+def test_fetch_lyrics_text_skips_already_cached(tmp_path: Path):
+    (tmp_path / lyrics.LYRICS_TEXT_CACHE_FILE).write_text(
+        json.dumps({"100": "уже загружено"}, ensure_ascii=False), encoding="utf-8"
+    )
+    client = FakeClient({"100": "новый текст"})
+
+    cache = lyrics.fetch_lyrics_text(client, ["100"], tmp_path, delay=0)
+
+    assert client.requested == [], "уже закэшированный трек не должен запрашиваться снова"
+    assert cache["100"] == "уже загружено"
+
+
+def test_fetch_lyrics_text_stores_none_for_missing_track_and_does_not_retry(tmp_path: Path):
+    # NotFoundError - окончательный ответ, а не сбой; None пишется в кэш как "текста нет",
+    # иначе такой трек перезапрашивался бы при каждом запуске.
+    client = FakeClient({})
+
+    lyrics.fetch_lyrics_text(client, ["100"], tmp_path, delay=0)
+    second_client = FakeClient({})
+    cache = lyrics.fetch_lyrics_text(second_client, ["100"], tmp_path, delay=0)
+
+    assert cache["100"] is None
+    assert second_client.requested == []
+
+
+def test_fetch_lyrics_text_stores_none_when_supplement_has_no_lyrics(tmp_path: Path):
+    client = FakeClient({"100": None})
+
+    cache = lyrics.fetch_lyrics_text(client, ["100"], tmp_path, delay=0)
+
+    assert cache["100"] is None