AI-powered music discovery for the web. Tuned In analyzes any webpage and recommends songs that match its mood, energy, and themes—all powered by on-device AI for complete privacy.
When you click "Generate," the extension injects Mozilla's Readability library into the active tab. Readability strips away navigation, ads, and boilerplate to extract the article's core text content—the same algorithm that powers Firefox's Reader View.
The extracted text is fed to Chrome's experimental Summarizer API, which runs Gemini Nano locally on your device. No data is sent to external servers.
For long pages, the text is chunked (configurable 1K–10K characters) and summarized in passes:
- Each chunk is summarized independently
- Chunk summaries are concatenated
- If the combined summary exceeds the model limit, it's summarized again into a final condensed form
The summarizer is instantiated once per session and reused—no repeated model loading.
The summary is passed back through the same on-device model with a structured prompt that requests:
- BPM (tempo): 60–180, mapped from the content's perceived energy
- Genres: 2–3 music genres matching the content's mood
The prompt explicitly constrains output to valid Spotify seed genres and prevents the model from suggesting content-based genres like "thriller" or "documentary."
Example output:
bpm: 110
genres: ["indie", "chill", "folk"]
AI-suggested genres are normalized to Spotify's supported seed genres via a mapping table. This handles synonyms (hip hop → hip-hop, r&b → r-n-b) and mood-to-genre conversions (romantic → r-n-b, aggressive → metal).
Invalid genres are filtered against Spotify's official seed genre list. If nothing maps, fallback genres (pop, chill) are used.
BPM is converted to Spotify audio feature targets:
| BPM Range | Energy | Description |
|---|---|---|
| < 80 | 0.2–0.4 | Slow, ambient |
| 80–100 | 0.3–0.5 | Chill, mellow |
| 100–120 | 0.4–0.6 | Moderate, groovy |
| 120–140 | 0.6–0.8 | Upbeat, driving |
| 140+ | 0.7–1.0 | Fast, intense |
Valence (musical positivity) is derived from genre keywords:
sad,melancholic→ 0.2aggressive,metal→ 0.3romantic,chill→ 0.6happy,party→ 0.8
The extension employs a multi-strategy approach to find the best match:
Strategy 1: Genre-Filtered Search
Queries Spotify's Search API with genre: filters:
genre:indie genre:chill
Multiple genre combinations are tried (single, dual, triple) with randomization for variety. Tempo keywords are occasionally added as a secondary signal, but deprioritized since they're unreliable in track metadata.
Strategy 2: Multi-Genre Text Search
Falls back to plain text queries combining genre names:
indie chill folk
Strategy 3: Playlist Mining
Searches for playlists matching the genre profile, then samples tracks from a random result. This surfaces tracks that wouldn't appear in direct searches.
Strategy 4: Simple Fallback
Single-genre search with no filters—ensures something always returns.
Since Spotify deprecated the Audio Features API for free-tier apps, tracks are scored using:
| Factor | Weight | Source |
|---|---|---|
| Genre relevance | 85% | Track/artist name contains genre keywords |
| Popularity | 7.5% | Spotify popularity metric (0–100) |
| Tempo relevance | 7.5% | Bonus-only; tempo keywords in track name |
Tracks are scored, sorted, then artist-diversified—taking the top track from each unique artist before filling the pool with remaining high-scorers. A random selection from this pool ensures variety across runs.
The dual-range slider sets min/max popularity bounds. If no tracks fall within the range, bounds expand by ±10 iteratively until matches are found or the full 0–100 range is used.
- 100% local AI processing via Gemini Nano
- No data collection or tracking
- No external API calls except Spotify (for track search only)
- Open source and auditable
- Chrome 138+ (Stable channel)
- Node.js & npm
- Spotify Developer account
git clone https://github.com/ClaytonWas/tuned-in.git
cd tuned-in
npm install
npm run build- Open
chrome://extensions/ - Enable "Developer mode"
- Click "Load unpacked" → select
distfolder - Deploy tuned-in-api backend
- Configure your origin trial token in
manifest.json
| Setting | Description |
|---|---|
| Full Text Mode | Process entire page vs. first chunk |
| Chunk Size | Characters per processing chunk (1K–10K) |
| Theme | Light / Dark |
| History Limit | Max saved recommendations (1–100) |
| Popularity Range | Min–max Spotify popularity filter |
| Show Scrollbar | Toggle scrollbar visibility |
Note: The Summarizer API requires Chrome 138+ and an origin trial token. See Chrome's AI documentation for setup details.