Architecture
Overview
Section titled “Overview”reklawdbox is a single binary that talks to your AI agent over stdio. It reads your Rekordbox database, fetches metadata from external services, analyzes audio, and exports XML. No runtime dependencies beyond an optional Python install for Essentia.
Data flow
Section titled “Data flow”Your AI agent (Claude Code / Claude Desktop / Codex) │ │ stdio ▼reklawdbox ├── Rekordbox master.db ──── read-only, encrypted ├── Local cache ──────────── enrichment, audio analysis, sessions ├── Discogs broker ───────── OAuth proxy for metadata lookups ├── Beatport ─────────────── metadata extraction (no API key needed) ├── Audio analysis ───────── tempo/key, rhythm/structure, dynamics, timbre └── XML output ───────────── Rekordbox-compatible export filesRekordbox database
Section titled “Rekordbox database”Rekordbox 6/7 encrypt master.db with SQLCipher. reklawdbox bundles everything needed to decrypt it and opens it read-only.
Key tables
Section titled “Key tables”| Table | What’s in it |
|---|---|
djmdContent |
Tracks (title, artist, genre, BPM, key, rating, comments) |
djmdPlaylist |
Playlists and folders |
djmdSongPlaylist |
Which tracks are in which playlists |
djmdArtist, djmdAlbum, djmdGenre, djmdKey, djmdLabel, djmdColor |
Lookup tables |
Discogs broker
Section titled “Discogs broker”Discogs requires OAuth for API access. A companion service on Cloudflare Workers handles this so reklawdbox never holds consumer secrets directly.
-
Ask the agent to look up any track on Discogs.
-
On first use, you get an authorization URL — open it in your browser and approve.
-
After that, lookups work automatically. Your session is stored locally in macOS Keychain.
The broker is open-source and self-hostable if you prefer. See Environment Variables for configuration.
Audio analysis
Section titled “Audio analysis”Two analysis backends are cached per track and reused while their schema, audio-file identity, and analyzer-specific inputs remain current. An audio-file identity change invalidates both backends. A Stratum schema change invalidates only Stratum, and an Essentia schema change invalidates only Essentia. A Rekordbox grid change also invalidates only Stratum.
Built-in (stratum-dsp) — BPM/key confidence, grid provenance and stability, decay, dub-stab and kick-pattern evidence, and structural sections in Rust. Always available, no setup needed.
Essentia (optional) — adds integrated loudness, danceability, rhythm and
onset evidence, brightness, and other spectral/timbral features via a Python
subprocess. It has no returned energy field. Transition and pool scoring
derive energy from danceability, integrated loudness, and onset rate only when
all three are present. Installed by reklawdbox setup.
Without complete Essentia energy inputs, scoring falls back to a BPM-based estimate; unavailable brightness/rhythm axes are excluded from the transition composite. Classification, transition scoring, pool scoring, audit, calibration, and audio-profile summaries each use only a subset of the returned evidence. See the canonical audio evidence and consumers table.
The same binary also works as a standalone CLI — no AI agent needed. Useful for batch operations:
| Command | What it does |
|---|---|
hydrate |
Batch enrichment + audio analysis |
analyze |
Audio analysis only |
backup |
Manage Rekordbox library backups |
read-tags / write-tags |
Read or write audio file tags |
extract-art / embed-art |
Cover art operations |
setup |
Install Essentia and configure MCP hosts |
See CLI Commands for details.
