Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.patchline.ai/llms.txt

Use this file to discover all available pages before exploring further.

Every track in your Patchline catalog gets deep audio analysis — BPM, key, mood, energy, valence, danceability, sonic genres, and instruments. This isn’t LLM-vibes-based-on-the-title. Patchline runs real audio analysis and writes the features back to your catalog.   5 AI credits per track. Runs automatically on import and upload.

​
What gets extracted

FeatureRangeWhat it measures
BPM40–220Beats per minute
KeyC, C♯, D … B (+ major/minor)Detected musical key
Modemajor / minorKey mode
Energy0.0–1.0Perceptual intensity (loud, fast, noisy = high)
Valence0.0–1.0Musical positivity (happy/cheerful = high; sad/angry = low)
Danceability0.0–1.0How suitable for dancing — based on tempo, rhythm stability, beat strength
Acousticness0.0–1.0Confidence track is acoustic
Instrumentalness0.0–1.0Likelihood of no vocals
Liveness0.0–1.0Likelihood of live audience
Speechiness0.0–1.0Spoken-word content presence
LoudnessdBOverall loudness (typically -20 to 0)
Time signatureintEstimated time signature (3, 4, 5, …)
Sonic genrestagsMulti-label genre tags (e.g., dreampop, lo-fi, bedroom-pop)
Sub-genrestagsFiner-grained tags
Mood tagstagsmelancholy, uplifting, aggressive, chill, etc.
Instrument tagstagsDetected instruments — electric guitar, synth pad, drum machine, …
Vocal presenceboolHas vocals or not
Vocal gendertagsDetected vocal gender(s)

​
Why it matters

Sonic features are the substrate for almost every downstream Patchline feature:

Playlist matching

find_playlists ranks Spotify playlists by sonic fit — that “fit” is the cosine similarity between your track’s features and the playlist’s average features.

Similar track discovery

Embeddings derived from sonic features power “tracks like this in your catalog” and “artists who sound like you.”

DSP pitch generation

Pitch copy references concrete features (“F minor, 124 BPM, dreampop-adjacent”) instead of vague descriptors.

A&R scout

Scout uses sonic embeddings to surface artists with sonically similar catalogs, not just genre-matching ones.

​
When it runs

TriggerLatency
Catalog import (URL)Queued immediately, completes in ~30–90 s
Catalog upload (file)Queued after upload completes
Manual re-analysis (catalog UI ⋯ menu)Same queue, ~30–90 s
Aria asks for itAria calls get_audio_features — cached if available, runs analysis if missing
You don’t need to start it manually. Sonic analysis is queued the moment a track lands in your catalog.

​
Two analysis layers

Patchline combines two complementary analysis layers:

​
Track features

The first layer extracts the music features you see in the catalog UI: BPM, key, mood, genre, instruments, vocal presence, energy, valence, danceability, and confidence signals.
  • Used by catalog filters, pitch generation, playlist matching, and Aria.
  • Usually completes in 30-90 seconds for a standard track.
  • Costs 5 credits per track when analysis runs.

​
Similarity features

The second layer creates similarity features that help Patchline answer questions like “what else sounds like this?” and “which artists are sonically close to this track?”
  • Used by similar-track discovery, Scout, and playlist fit.
  • Returned only when the track has enough audio context for a reliable comparison.
  • Combined with metadata and audience context when ranking matches.
You don’t pick between layers. Patchline uses the best available analysis for each track and surfaces the combined result.

​
How to view a track’s analysis

​
In the dashboard

Sidebar → Catalog → click any track row → Audio features panel on the right. Each numeric feature renders as a horizontal bar (0–1 scale) with the exact value. BPM and key render as labels. Tag lists (genres, moods, instruments) render as chips.

​
Through Aria

​
Through the MCP server

get_audio_features returns the raw numeric + tag features. get_song_intelligence returns features plus ISRC, streaming stats, playlist matches, and collaborator info — a one-shot deep dive.

​
Examples

​
Example 1 — Find your most energetic track

Aria: That's "Run Hot" — energy 0.84, BPM 142, key A minor.
It's also the most danceable (0.79). Want me to find playlists matching
that profile?

​
Example 2 — Use features in a pitch

Aria: "Sleeper" is a 76 BPM, F# minor dreampop ballad with low energy
(0.31), low valence (0.21), and high acousticness (0.78). The track sits
between Mazzy Star and Phoebe Bridgers sonically. Here's a draft pitch
for late-night editorial curators... [continues]

​
Example 3 — Verify a guess

Aria: It's 112 BPM, key D major. Time signature 4/4. Want the full
feature breakdown?

​
Pricing

ActionCredit cost
Sonic analysis per track5 credits
Re-analysis (re-runs the same pipelines)5 credits
Reading cached features via get_audio_features0 credits (cached)
get_song_intelligence (features + streaming + matches)1 credit
A typical Free user runs sonic analysis on a few tracks/month (~30 credits). A Pro user with active import flow might run 20–50/mo (~100–250 credits).

​
Limitations

  • Metadata-only assets can’t be analyzed. If you imported a track via URL but the streaming source can’t fetch the audio, the track stays at “queued.” Solution: upload the audio directly.
  • Analysis is best-effort. Very short tracks (under 30s), heavily noisy recordings, or speech-dominated tracks can return partial features. Confidence indicators are exposed in the API response.
  • Re-analysis costs credits. If you re-upload a master, the new version gets re-analyzed. The old version’s features remain in the catalog but flagged as superseded.

​
Privacy

  • Uploaded audio is stored in encrypted, access-controlled cloud storage.
  • Analysis runs only to provide Patchline features for your workspace.
  • Patchline does not use your audio to train models.
  • Short-lived access is used for analysis and playback; implementation details are intentionally kept out of public docs.

​
FAQ

Tags explain the track in human terms: mood, genre, instruments, energy, and vocal presence. Similarity features compare the track to other tracks and artists. Patchline combines both because pitches, playlists, and A&R workflows need explanation and comparison.
Mostly yes. BPM is reliable above ~85% accuracy on standard musical content. Key detection drops in confidence for atonal, ambient, or heavily modulating tracks. The API exposes confidence; the UI shows a ?? icon when confidence is below threshold.
Patchline uses a multi-label taxonomy aligned with common DSP and music-industry genre language. Some tracks receive multiple genres or sub-genres because that is more useful than forcing one label.
30–90 seconds for a typical 3-minute song. Longer tracks and large batches can take more time. The catalog UI updates when results land.
Yes for individual tracks (catalog row → ⋯ → Export). Bulk CSV export is Enterprise-only via the REST API.
No. Audio is used to provide analysis features for your workspace, not to train models.