Everything you need torun your whole collection

Three libraries, pure-Go analysis, AI tagging, similarity discovery, audio-quality checks, DAW kit generation, DJ set curation, MIDI pattern generation and voice synthesis — in one signed macOS binary.

Three Indexes, One App

Samples, finished tracks and MIDI files each get their own tables, scan pipeline and similarity index — switched with a single toggle that the browser, explore and tag pages all follow.

Sample library

Recursive worker-pool scanning with content hashing for incremental re-scan, and a virtualized table that stays smooth across very large collections.

Track library

A separate index for full records, with a 1–10 energy score, structural cue detection (intro, drop, breakdown, outro), editable cue points, artwork and a preview offset that skips the intro.

MIDI library

An index over your .mid / .midi packs with real content analysis: pitch-class key detection, drum / melodic / chords / mixed classification, polyphony and chord-progression summaries.

Drag & drop, both ways

Drop files or folders onto the window to import into whichever library you are looking at; drag samples straight out into Ableton, Logic, Pro Tools or Finder via a real macOS drag session with a waveform thumbnail.

Organize

Favorites, per-item memos, folder color dots, a colored tag tree, folder path-tree navigation and full-text search — in all three libraries.

Third-party import

Auto-detects Splice, Loopcloud, Ableton, Logic, Apple Loops, Native Instruments and MPC default locations.

Pure-Go DSP, No Services

Everything is computed in-process the moment a file lands. No Python sidecar, no upload, no per-file API cost.

Tonality detection

HPCP chroma front-end correlated against Krumhansl-Schmuckler key profiles across all 24 major/minor rotations, with a bass-note tonic tie-breaker. Returns key, scale and confidence.

Tempo

Spectral-flux onset envelope → autocorrelation → comb scoring, with half/double-time correction, loop-grid snapping and sub-BPM interpolation.

Timbre

Spectral centroid (brightness), rolloff, flatness, bandwidth, spectral contrast and 13 MFCCs — means and variances.

Envelope

Attack time, RMS, zero-crossing rate and a loop / one-shot heuristic.

DJ analysis

Tracks additionally get a 1–10 energy score blended from RMS, brightness, tempo and zero-crossing rate, plus structural cue points you can correct by hand.

Perceptual embeddings

48-dimensional vectors grouped into timbre, rhythm and pitch, powering weighted-cosine similarity search with the group weights exposed as sliders.

Spectrogram & Fake-Lossless Detection

A Spek-style time/frequency heat map rendered in Go at the file’s native sample rate over its full length — not the analysis excerpt, and not resampled down to 22 kHz, because that would hide exactly the band you need to see.

Inline or full-size

Toggle the waveform for a spectrogram without losing transport or seek, or open it as a full-size dialog with a quality chip.

Bitrate estimation

The lowpass cutoff is read off the long-term average spectrum rather than a max-hold, so a single click or edit point cannot smear broadband energy across the shelf and hide it.

Conservative by design

The bitrate table is only consulted for a genuinely sharp shelf. Plenty of real masters fade out by 18 kHz, and labelling those “lossy” would flag most of a library.

Fake-lossless flag

Reserved for a lossless container with a very steep drop and at least half a minute of audio — a short one-shot is legitimately dark up top.

Categorized in Plain Language

Bring your own key, or use the Claude CLI you already pay for. Nothing but filenames, metadata and computed features is sent — never audio.

Any provider

Anthropic, OpenAI or Gemini API keys, or the authenticated claude CLI — billed against your own Claude plan with no API key at all.

Two models, two budgets

Tagging runs on a cheap classification model; the tool-using agents get an independent, reasoning-grade one. Making the agents smarter does not change your per-sample tagging cost.

A real taxonomy

Nine dimensions — category, genre, character, mood, effects, playing technique, song structure, gear, type — with aliases, driving both the model’s vocabulary and the UI tag tree. MIDI gets its own taxonomy.

Heuristic fallback

With no provider configured, a deterministic filename-plus-feature heuristic still tags everything, entirely offline.

Natural-language selection

Type “select all dark 808s” and it resolves to a filter constrained to your library’s real vocabulary, for samples or for tracks.

Find the Right File

Similarity, reference matching and visual exploration, across whichever library you are in.

Similarity search

Weighted “sounds like” nearest-neighbour over the embeddings, tunable across timbre, rhythm and pitch, with optional key-compatibility and same-category constraints.

Query by file

Drop in any external file — even one you do not own — and find the closest matches you already have. It is analyzed without being imported.

Match Track

Analyze a reference and surface samples that fit by compatible key (Camelot / circle-of-fifths), BPM window and category.

Explore views

A 3D Sound Galaxy for samples plus PCA maps for tracks and MIDI, each with a dashboard, key wheel, taxonomy panel and feature-insight charts.

Deep filtering

Key and scale, BPM range, brightness, duration, energy, loop vs one-shot, favorites, tags, path segments and free text — combined.

Duplicates and same-song

Group samples by content hash to reclaim disk space. For records, a separate song-identity check catches the same song stored as two different files, and repairs sets that already contain both.

Preview, Trim and Convert

Audio is streamed to the UI with real range requests and an allowlist check, never base64 data URLs.

Waveform player

Web Audio preview with a canvas waveform (mono or stereo), transport, click-to-seek, loop toggle and a spectrogram view.

Sample-accurate trim

Cut a WAV to a selection in place or as a copy, preserving channels, sample rate and bit depth, with the index kept in sync.

MIDI playback

Click any MIDI file to hear it through a built-in polyphonic synth, with a piano-roll bar showing exact unquantized note events.

WAV export

Convert a selection to WAV at a target sample rate, at 16 or 24-bit.

From Vault to Session

Both DAW writers inject into real exported reference files rather than hand-authoring their schemas, which is why the results actually load.

Ableton Live

Drum Rack (.adg) and Live Set (.als) generation from any selection, on 8 to 64 pads.

Maschine 2

Group (.mxgrp) and project (.mxprj) generation, with optional portable sample copying for a self-contained kit.

Curated pad layout

The default layout reads each pad’s role from the taxonomy tags, places one representative per role in canonical kit order, then fills spare pads with the variant farthest from what that role already has. Rank order and group-by-type remain available.

Saved kits

Kits are named, editable entities with export history. Leave one live and the pads re-resolve from its saved filter on every export; freeze the layout and it re-exports identically even after a re-scan renumbers the library.

Kit agent

Describe the kit you want and an agent searches your library for it, then drops a validated pad plan into the same editable grid you would have filled by hand. Empty pads are completed by the deterministic curator.

MIDI patterns

Nine genre templates (techno, psytrance, house, tech house, minimal, DnB, breakbeat, hip-hop, trap) with swing, humanize, variation and seed — exported as .mid, or as an .als whose drum rack is bound to your own samples. Existing .mid files can be imported onto the grid.

Curate and Export a Set

Everything here works on the track index, never on samples.

Playlists with chapters

Named chapters, free ordering, per-item notes and colors, plus a node-graph view of the set.

Magic Sort

A greedy chain over harmonic compatibility, BPM proximity and energy progression, with a selectable energy shape (rising, falling or arc), scoped to the whole set or one chapter.

Recommend, complete, vary

Ask for the next track from the tail, fill a set out to length by chaining or by set average, or clone a set into a new one with similar-track substitutions.

AI set generation

Give it a brief and watch the step log as it searches, checks key compatibility and pulls neighbours. The result is a reviewable draft — named chapters, ordered tracks, per-track notes — that only becomes a set when you commit it.

No duplicate songs

Because the index is path-keyed, one record kept as two files is two rows. Generation, completion and variation all de-duplicate on song identity rather than row id, and existing sets can be repaired in one click.

USB export

Export a set in order, copied verbatim or re-encoded to WAV, with optional numeric filename prefixes and per-chapter subfolders.

Generate What You Don’t Have

Rendered clips land in the library as ordinary samples and go through the same analysis and tagging as anything you scanned.

Text-to-speech takes

ElevenLabs voice renders with control over stability, similarity, style and speed, batched as variations and written out as mono 16-bit WAVs.

Prompted sound effects

Describe a sound, set a length and a prompt-adherence level, and file the result straight into the vault.

Clean output

Optional peak normalization to −1 dBFS and silence trimming, written atomically so a failed render never leaves a partial file behind.

Streaming sync

Read-only Spotify and SoundCloud sync mirrors your own likes and playlists into the track index. Nothing is ever written back to the platform.

Local by Default

The index, the audio and all DSP stay on your machine, and there is no telemetry. Everything that touches the network is a feature you switch on, with your own credentials.

Single signed binary

One notarized macOS app with an embedded UI and a local SQLite index — no Docker, no sidecar, no account.

No telemetry

No analytics, no crash pings, no usage reporting. Data lives under ~/Library/Application Support/AudioVault.

Opt-in integrations

AI tagging, license activation, the updater, artwork lookup, voice generation and streaming sync are the only things that ever leave your Mac — each one off until you configure it.

Trial and lifetime license

A 14-day trial that activates itself on first launch with no signup or card, then a one-time purchase verified with Ed25519 signatures, plus an in-app self-updater.

Specifications

Audio formats
WAV, AIFF, FLAC, OGG Vorbis, MP3 and MP4/M4A/AAC (AAC decoding uses the system transcoder)
MIDI formats
.mid
Key detection
24 Krumhansl-Schmuckler profiles (12 major + 12 minor)
Embeddings
48 dimensions — timbre, rhythm and pitch groups
Export targets
.wav (16/24-bit), .adg, .als, .mxgrp, .mxprj, .mid, ordered set folders
Pad grids
8, 16, 32 or 64 pads
Pattern genres
9 built-in templates
Tag taxonomy
9 dimensions, ~237 terms with aliases
Requirements
macOS 12 or newer, Apple Silicon or Intel
Index
Local SQLite (WAL), pure Go — no CGo, no server