Next Track

What should I play next? Next Track suggests your next tune by how tracks actually sound, filtered to a mixable BPM and a harmonically compatible key.

Built for the Appwrite hackathon (Oct 2026) on Appwrite VectorsDB, which does the recommending: every track is stored as a 512-dimension CLAP audio embedding, and "what's next" is a cosine-similarity vector query combined with BPM and Camelot-key filters.

Work in progress. See docs/PLAN.md for the design and schedule.

Indexing your crate

# 1. Analyse locally (audio never leaves your machine)
cd ingest
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python analyse.py ~/Music/crate --range 1-100   # appends to ../out/tracks.jsonl
python analyse.py ~/Music/crate --range 101-200 # already-analysed tracks are skipped

# 2. Upload fingerprints + metadata to VectorsDB
cp ../.env.example ../.env   # fill in project ID, API key, IDs
npm install
npm run upload

Before uploading, create in the Appwrite Console a VectorsDB database nexttrack with a collection tracks (dimension 512) and an hnsw_cosine index on embeddings.

Running the app

cd web
npm install
npm run dev

The app reads the same repo-root .env as the ingest scripts. Without APPWRITE_PROJECT_ID and APPWRITE_API_KEY it shows a generated demo crate, so the UI works on a fresh clone.

Pick a record from the crate, and suggestions orbit the deck on a Camelot wheel: each sits in its key's slice, and the closer it is to the deck, the more alike it sounds. Drag one onto the deck (or click it) to play it next.

Deploys to Appwrite Sites with the Node adapter (npm run build, output in web/build). Set the same environment variables in the site's settings.