Why

TikTok will not let you save the sound. "Favorites" are bookmarks that vanish the moment a video is deleted, a track is taken down, or an account goes private. Screen-recording is lossy, the sketchy mp3 sites are malware, and the link you texted yourself rots in your notes.

Sound Cache fixes that permanently. You save a sound once and it lands as a real, tagged audio file in a local folder you own, complete with the title, artist, artwork, transcript, and example videos, so your FYP's greatest hits outlive platform churn. Built for the editors and creators who need the trends offline and organized.

How it works

1. On TikTok, tap the sound (the spinning disc) to open its sound page.
2. Tap Share ↗, then More •••, then Save to Sound Cache.
3. The share sheet hands off the link (never your audio) to a tiny relay keyed to your personal pair code.
4. Your desktop pulls the link, downloads the real audio with your own logged-in session, and files it with artwork plus a transcript.

One tap on your phone, and a fully-tagged file appears on your computer. Offline. Unbothered. The relay only ever holds a link on its way to you, with no account, no files, and no tracking.

Setup:

1. Download and install the Sound Cache mac app from Releases
2. Click the settings button in the lower left, then click Pair iPhone
3. Scan the QR code and install the IOS Shortcut.

Caching from TikTok REQUIRES an active logged in session on your computer

1. Click Settings in the Sound Cache App
2. Clock Connect TikTok
3. Login here with any credental. You can use any credental, like a burner account as the functionality is limited to an active web session.]

Features

• File-native vault. A metadata.json per sound is the source of truth; SQLite is a disposable, rebuildable full-text cache. Everything stays browsable in Finder or on a NAS even if the app never runs.
• Search by anything. Title, artist, music ID, tags, spoken phrase (transcript), usage and popularity, status, or local-media state.
• Rich ingest. Multi-platform (TikTok, Instagram, YouTube) via yt-dlp with a Playwright capture fallback; pulls artwork, popularity, transcript, and example videos. A shared TikTok video resolves back to its underlying sound so you cache the clean full track, not the trimmed clip.
• GPU transcripts. On-device speech-to-text via Whisper, accelerated on your hardware: Apple's MLX runs it on the Apple-Silicon GPU, NVIDIA cards get CUDA automatically, and everything else falls back to a fast CPU path. A background enrichment cycle quietly fills in anything still missing a transcript.
• Spotify links. When a sound maps to a commercial track, Sound Cache captures the Spotify link so you can jump straight to the original.
• Editor-friendly. Drag a sound out of the window to drop the audio file straight into Premiere, Resolve, or CapCut. Favorites, sorting bins, duplicate review, and archive-health coverage are all built in.
• Private pairing. A one-time pair code links your phone's share sheet to your desktop, and the lower-left badge confirms the connection at a glance.
• Portable by design. Sound names with kaomoji, emoji, or exotic Unicode are sanitized for any filesystem (byte-accurate length caps, NFC normalization, emoji sequences preserved) so your vault copies cleanly to a NAS or another drive.

Privacy and security

Local-first means what it says. Your collection lives on your machine, not on a server. There is no account, so there is nothing to leak.

• The relay only ever holds a link briefly on its way to your desktop (24-hour TTL), never your files and never a profile of you, and you can stop it in one click.
• Submitted URLs are validated (http and https only; private, reserved, and cloud-metadata hosts rejected) at the relay and again at the desktop before any fetch. It is SSRF-hardened, with a safe-redirect handler.
• All request fields are length-bounded, per-pair-code flood caps protect your inbox, and logs redact URLs and never store free-form notes server-side.
• Opt-in leaderboard telemetry is anonymized (a sound id, title, and platform only): no account, no device secret, no paths.

Full Privacy Policy and Terms of Use, including the anti-piracy notice, live on the site.

Tech

Python 3.12  ·  PySide6 (Qt6) desktop  ·  FastAPI relay on Vercel with Neon Postgres  ·  SQLite FTS5  ·  yt-dlp and Playwright for ingest  ·  faster-whisper and MLX for transcription.

Run it (dev)

# 1. External tools. TikTok sound capture drives a real browser via Playwright;
#    yt-dlp and ffmpeg handle download and transcode.
brew install node ffmpeg yt-dlp           # macOS (use your package manager elsewhere)
npm install                               # installs Playwright into ./node_modules
npx playwright install chromium           # one-time browser download

# 2. Python app.
python -m venv ~/venvs/sound-cache && source ~/venvs/sound-cache/bin/activate
pip install -e ".[gui,ingest]"            # GUI + on-device transcription + yt-dlp
sound-vault                               # launch the desktop app
# in-app: Settings -> Create pairing code -> Pair iPhone -> Connect TikTok

You need an active TikTok session. TikTok serves a sound's audio only to a logged-in browser, so the first run walks you through Connect TikTok, a one-time login the app keeps locally on your machine and never uploads. Without Node, Playwright, and Chromium the app still runs, but TikTok sound capture is disabled, and the in-app prompts tell you exactly what is missing.

Apple Silicon note. The transcription engines ship as native arm64 wheels. Launch the app natively (not under Rosetta) so the GPU path can load; the bundled launcher forces the arm64 slice for you, and the app warns you if it ever finds itself translated.

Tests: pytest -q (set SOUND_VAULT_DISABLE_RELAY_POLL=1 SOUND_VAULT_DISABLE_TRANSCRIBE=1 for a fast offline run).

License

MIT. Sound Cache is a tool for personal organization of sounds you are entitled to use; please respect creators' rights and platform terms.