Sound Forge Alchemy
v4.1.0 — AI-powered audio pipeline with multi-LLM agents, stem separation, and DJ/DAW tools
NOT FOR COMMERCIAL USE — USE AT YOUR OWN RISK. This is a personal creative project with no warranty, no support, and no guarantees. This is mostly for vibe guys.
Sound Forge Alchemy (SFA) is a production-grade audio toolkit built on Elixir/OTP and Phoenix LiveView. Import tracks from Spotify, separate stems locally via Demucs or in the cloud via lalal.ai, analyze audio features with librosa, query a multi-LLM agent system for music intelligence, and control it all in real time through MIDI/OSC hardware.
Main library view: 67-track collection with album art, sidebar navigation (Library, Playlists, Browse, Studio), and Spotify player integration.
Quick Navigation
| Section | Description |
|---|---|
| Architecture | System design, OTP tree, data model |
| Guides | Quickstart, installation, configuration, deployment |
| Features | Import pipeline, stem separation, analysis, DJ/DAW, AI agents, admin |
| API Reference | REST endpoints and WebSocket channels |
| Changelog | Release history |
| Contributing | Development guidelines |
Feature Highlights
Authentication page supporting both magic link (email-only) and traditional password login flows.
- Spotify Import — Paste any Spotify URL (track, album, playlist). Metadata fetched via OAuth2 client credentials; audio downloaded by
spotdl. - Dual Stem Engine — Local Demucs (htdemucs, htdemucs_ft, htdemucs_6s, mdx_extra) or cloud lalal.ai with 9+ stem types and 60-second preview.
- Audio Analysis — librosa-powered feature extraction: tempo, key, energy, MFCC, chroma, spectral centroid. Rendered as D3.js visualizations.
- AI Agent System — Six specialist agents (TrackAnalysis, MixPlanning, StemIntelligence, CuePoint, Mastering, Library) orchestrated by a routing Orchestrator. Pluggable LLM backends: Anthropic, OpenAI, Google Gemini, Ollama, Azure OpenAI.
- DJ / DAW Tools — Two-deck DJ mixer with BPM sync, loop controls, and EQ; multi-track DAW editor with MIDI export and per-stem mute/solo.
- MIDI/OSC Control — Hardware controller mapping via
midiex; Open Sound Control server for DAW integration. - Admin Dashboard — Role-based access (
user,admin,platform_admin) with audit logs, user management, and analytics. - Real-Time Pipeline — Oban background jobs, Phoenix PubSub, LiveView streams — no page refresh required.
Recent Changelog
v4.1.0 (2026-02-25)
- Azure Container Apps production deployment
- SSL termination, DAW fixes, comprehensive documentation
v4.0.0 (2026-02)
- lalal.ai full cloud stem separation integration (82 files, +12,398 lines)
- 9+ lalal.ai stem types, 60-second preview, quota management
v3.0.0 (2026-01)
- Audio analysis expansion: MFCC, chroma, spectral, 5 D3.js visualizations
- 24 files, +4,712 lines
Getting Started
Registration page: create an account to begin importing tracks and building your library.
For setup instructions see the Quickstart Guide.
Repository Links
About
Sound Forge Alchemy is a project by Jeremiah Pegues. The concept, original architecture, and feature design are entirely his. This repository is a complete port of his original React/Node.js prototype to Elixir/Phoenix.
Built with Claude Code
The port was done almost entirely using Claude Code (Anthropic). Custom skills /elixir-architect and /ralph were instrumental — handling OTP supervision design, Phoenix context boundaries, PRD-driven story decomposition, and TDD enforcement across all 707 tests.