Features

Everything a prepared library needs, without the cloud

01 · Analysis

Native analysis, seconds per track

<4 s warm per track<0.5 s waveform peaksNeural Engine + vDSP

Rhythm, structure, loudness, pace, and key come from Apple's native audio analysis frameworks, accelerated on the Neural Engine with vDSP. Warm-cache analysis targets under 4 seconds per track; waveform peaks under half a second.

Per-bar DSP detail — sub-bass energy, spectral features — is computed locally with Accelerate. This is the same machinery your OS uses, with no cloud round-trip. Semantic labels like build and drop are inferred in Metamix from that geometry and evidence.

MusicUnderstanding + Accelerate on Apple Silicon

Screenshot coming soon

Native analysis, seconds per track

02 · Cue engine

Role-first cues on the grid you trust

8/16-bar quantizeFake-drop demotionmanual_rb guard

The Go cue engine detects builds, drops, and breakdowns from sub-bass energy, pace, and instrument activity — then snaps every cue to the 8/16-bar hypermeter on top of your Rekordbox beatgrid. Fake drops that collapse after entry get demoted automatically.

Hot cues, memory cues with spacing and dedupe, and per-role ordering are all validated before a single write. No valid grid, no cues. Tracks with manual Rekordbox cue edits are locked out of automated writes until you reset authority yourself.

drop_main · drop1 · memory landmarks · manual_rb guard

Screenshot coming soon

Role-first cues on the grid you trust

03 · Tagging

Genre and mood tags you can interrogate

Every tag explained512-d embeddingsLocal models only

A local model (AllMusicCaps) writes genre, subgenre, and mood tags, plus 512-dimensional embeddings for similarity.

Every tag carries its rule, evidence, and target — tag review shows you exactly why a track got tagged, and you can approve or reject before anything is applied.

Rekordbox MyTag categories map to customizable mix dimensions — Role, Components, Groove, Atmosphere — so CDJ Track Filter finds compatible rhythm, vocality, and headroom without scrolling genre soup.

Explainable tags · CDJ MyTags · 512-d embeddings · local models only

Screenshot coming soon

Genre and mood tags you can interrogate

04 · Discovery

Browse the whole archive like a streaming service

p95 < 150 ms facetsp95 < 100 ms search20k tracks

Browse by genre, BPM, key, energy, era, and tag. Full-text search across track, artist, album, and label. Similarity pivots from any track, artist, or album.

Queries target p95 under 150 ms on a 20,000-track library — fast enough to explore, not just to find. Your Beets or folder archive stays the source of truth; discovery never requires a cloud catalog.

FTS5 search · faceted browse · similarity pivots

Screenshot coming soon

Browse the whole archive like a streaming service

05 · Pipeline

Imports that survive real life

Resumable batchesMIK handoff optionalExplicit user gates

Select tracks from your archive, batch them, and import them into Rekordbox.

Batch jobs are durable and resumable — quit, crash, or close the laptop and the queue picks up where it left off. Mixed In Key exports feed the pipeline when you use them; final Rekordbox imports wait at explicit user gates.

Every stage is persisted, retried with backoff, and reviewed before it touches your library. Metamix orchestrates — it does not silently advance past an external Rekordbox or MIK step.

Resumable batches · optional MIK handoff · user gates

Screenshot coming soon

Imports that survive real life

Apple Silicon Performance

Real-world batch analysis throughput on M-series chips using native Accelerate/vDSP and Apple Neural Engine.

Full benchmark report
Reference Baseline Apple MusicUnderstanding + Accelerate/vDSP on macOS 27+
Local-first · Zero cloud audio upload
Apple Silicon Hardware Benchmarks
Hardware ModelThroughputSpeedupANE Latency

What Metamix doesn't do

  • · No cloud processing — analysis runs on your Mac only
  • · No pre-macOS 27 or Windows support
  • · No music downloads or streaming playback
  • · It doesn't replace Rekordbox's analysis — it builds on it
  • · Won't replace MIK's key detection

See it work on your library

Join the waitlist for early access and the locked-in launch price.

Join the waitlist