Aria

Music curation and playlist generation powered by AI

A service that creates personalized playlists based on seed tracks, using music metadata APIs, vector similarity matching, and LLM-powered explanations. An aria is a self-contained melody — and that's exactly what we create for you.

Domain: aria.grove.place Status: ⏸️ On Hold (as of January 2026)

Internal codename: GroveMusic — used in code, infrastructure, and deployments

Follow development progress: autumnsgrove.com/blog

---

Project Status

This project is currently on hold due to challenges in the legal and technical landscape for music metadata at scale.

Why the pause:

• Legal developments around large-scale music dataset availability (see SpotifyDataset_LegalResearch_2026-01.md)
• Gap between legal alternatives (~35M tracks) and the scale needed for the original vision (~256M tracks)
• Need to reassess approach and data sources before proceeding

The vision lives on: The idea of intelligent, AI-powered music curation remains compelling and viable. Aria may return when:

• Legal music metadata sources reach sufficient scale
• Alternative approaches to audio feature extraction mature
• The legal landscape around music data becomes clearer

The codebase, architecture, and research remain here as a foundation for future exploration.

---

Features

• Seed-based playlist generation - Enter a song, get a cohesive playlist that shares "musical DNA"
• Multi-source metadata - Combines Last.fm tags, MusicBrainz data, and vector embeddings
• AI-powered explanations - Each track comes with a reason why it fits your playlist
• Smart curation - Balances popular tracks, deep cuts, and hidden gems
• Credit system - Fair usage with Free, Basic, and Pro tiers

Tech Stack

Layer	Technology
Frontend	SvelteKit
Backend	Cloudflare Workers
Stateful Processing	Durable Objects
Database	Cloudflare D1
Caching	Cloudflare KV
Vector Storage	Cloudflare Vectorize
Object Storage	Cloudflare R2
Auth	Google OAuth 2.0

Development

Prerequisites

• Node.js 18+
• pnpm
• Cloudflare account
• Last.fm API key
• Google OAuth credentials

Setup

# Install dependencies
pnpm install

# Copy secrets template
cp secrets_template.json secrets.json
# Edit secrets.json with your API keys

# Create Cloudflare resources
wrangler d1 create grovemusic-db
wrangler kv namespace create CACHE
wrangler kv namespace create SESSIONS
wrangler kv namespace create RATE_LIMITS
wrangler kv namespace create CONFIG

# Apply database schema
wrangler d1 execute grovemusic-db --file=schema.sql

# Run locally
pnpm dev

Project Structure

src/
├── lib/
│   ├── components/     # Svelte components
│   │   ├── generator/  # Seed input, config, progress
│   │   ├── playlist/   # Playlist display, export
│   │   └── shared/     # Common UI components
│   ├── server/         # Backend logic
│   │   ├── api/        # API handlers
│   │   ├── pipeline/   # Pipeline workers
│   │   ├── services/   # External API clients
│   │   └── db/         # Database queries
│   └── types/          # TypeScript interfaces
└── routes/             # SvelteKit pages

Pipeline Overview

Input → Resolve → Enrich → Generate → Score → Curate → Explain → Output
         │          │         │         │        │         │
    MusicBrainz  Last.fm   Similar   Multi-   Balance   LLM
    lookup       tags      tracks    dimension categories explanations
                           Vector    scoring
                           search

Documentation

• grovemusic-spec.md - Full project specification
• AgentUsage/ - Development workflow guides

License

Private - Part of the Grove ecosystem