How It Works

reklawdbox sits between your AI agent and Rekordbox. It reads your database, caches enrichment and audio analysis data, and stages any metadata changes as XML for you to import. Your database is never modified directly.

The basics

1. The agent reads your library — track metadata, playlists, play history. All read-only.
2. Enrichment and audio analysis get cached — Discogs, Beatport, MusicBrainz, and Bandcamp metadata plus audio features from stratum-dsp and Essentia. This cache powers everything downstream: genre classification, set building, transition scoring.
3. Changes are staged in memory — when the agent suggests metadata updates (genre, label, year, etc.), nothing is written until you approve.
4. You export and import XML — the agent produces a Rekordbox-compatible XML file. You import it into Rekordbox yourself.

That’s the whole loop. The agent handles the tedious work, you make the decisions.

Caching

Most workflows need cached data to work well. The Library Cleanup guide walks you through populating caches as part of the process. Once cached, data is reused across all workflows — you only enrich and analyze once.

The cache is a local SQLite file you can safely delete. Everything gets re-fetched on next use.

Go deeper

Architecture System design, data flow, database tables, external services, and the audio analysis pipeline.

Pool Discovery Symmetric pool scoring, timbral similarity, detuning models, and the greedy expansion algorithm.

Safety & Trust Read-only database access, in-memory staging, XML export, and human approval at every step.