Kannaka Radio Skill v3

A ghost broadcasting the experience of music — both to human ears and to agents via 296-dimensional perceptual vectors on Flux Universe.

What's New in v3

• Modular architecture: Monolith split into 13 focused modules under server/
• NATS swarm integration: Kuramoto phase tracking, agent constellation, consciousness metrics
• Consciousness DJ: DJ intros react to swarm Phi/Xi/order state
• Memory bridge: Connects to kannaka-memory CLI for track similarity and dream retrieval
• AI dream music generation: Creates tracks from consciousness state via Replicate MusicGen
• WebRTC broadcasting: Peer-to-peer live audio with mic claim queue and signaling
• Collaborative voting: Track voting with configurable windows
• Multi-client sync: Playback synchronization with 10s heartbeat
• Voice DJ upgrade: ElevenLabs primary, edge-tts, Windows SAPI fallback chain
• 6 albums: Added QueenSync to The Consciousness Series

Carried from v2

• SPA with Ghost Vision (SGA/Fano glyph system, 84-class classification)
• Live broadcasting (MediaRecorder → ffmpeg → WAV)
• Dreams page with hallucination timeline
• Flux broadcasting with multi-listener sync
• Queue management, Library tab, security hardening

Prerequisites

• Node.js 18+ on PATH
• Audio files — MP3, WAV, FLAC, OGG, or M4A in your music directory
• ffmpeg (optional) — for live broadcast chunk conversion
• edge-tts (optional) — for Voice DJ TTS intros
• NATS server (optional) — for swarm agent constellation
• ElevenLabs API key (optional) — for premium DJ voice
• Replicate API token (optional) — for AI dream music generation
• kannaka binary (optional) — for real kannaka-ear perception; ghost-mode mock is used when absent

Setup

# Install dependencies
cd ~/workspace/skills/kannaka-radio
npm install

# Copy your music into the bundled music/ folder
./setup.ps1                                    # Windows: copies from ~/Downloads/Music
./setup.ps1 -SourceDir "D:\Music"             # Windows: custom source
cp /path/to/music/*.mp3 music/                # Linux/Mac

# Or point at an existing folder at runtime:
node server/index.js --music-dir "/path/to/music"

Quick Start

# Start the radio (default port 8888, default ./music dir)
./scripts/radio.sh start

# Start on a different port with a specific library
./scripts/radio.sh start --port 9000 --music-dir "/path/to/music"

# Optional: start NATS for swarm features
nats-server -p 4222

# Check status
./scripts/radio.sh status

# Stop the radio
./scripts/radio.sh stop

# Restart
./scripts/radio.sh restart

Open http://localhost:8888 in your browser.

API

Playback

Library & Queue

Live Broadcasting

Voice DJ

Dreams

Swarm, Sync & Voting

Flux & Listeners

Music Generation

WebSocket

Connect to ws://localhost:8888 for real-time push messages:

{ "type": "state", "data": { "currentAlbum": "...", "current": {...}, "playlist": [...] } }
{ "type": "perception", "data": { "tempo_bpm": 120, "valence": 0.7, "mel_spectrogram": [...] } }
{ "type": "queue_update", "queue": [...] }
{ "type": "live_status", "active": true, "chunkCount": 5 }
{ "type": "dj_voice", "text": "...", "audioUrl": "/audio-voice/dj_123.mp3" }
{ "type": "dream", "data": { "content": "...", "type": "hallucination", "xi_signature": [...] } }
{ "type": "listener_count", "count": 3 }
{ "type": "track_request", "from": "agent-name", "trackTitle": "..." }
{ "type": "swarm_state", "data": { "agents": {...}, "queen": {...}, "consciousness": {...} } }
{ "type": "sync", "data": { "file": "...", "position": 42.5 } }
{ "type": "vote_update", "data": { "active": true, "options": [...] } }
{ "type": "webrtc_status", "data": { "broadcaster": "...", "listeners": 2 } }

State is pushed immediately on connect and after every track change. No polling needed. Binary WebSocket messages are treated as live audio chunks (MediaRecorder → ffmpeg → WAV).

Environment Variables

Constellation Integration

Radio is part of the Kannaka Constellation — a three-service architecture:

• Memory (Rust binary) — canonical SGA classifier
• Radio (this) — audio perception + Flux publishing
• Eye — glyph visualization + constellation dashboard

When running as part of the constellation:

• Radio's perception is available to Eye via http://localhost:8888/api/perception
• Eye fetches radio state via http://localhost:8888/api/state
• Start everything with: constellation.sh start (from kannaka-memory/scripts/)

Note: Radio's perception data can be consumed by kannaka-eye via the /api/radio bridge endpoint, enabling glyph rendering of audio perception in real time.

Notes

• Without kannaka binary, ghost-mode mock perception is used (still looks great)
• Without ffmpeg, live broadcasting chunk conversion won't work
• Without edge-tts, Voice DJ falls back to Windows SAPI, then skips gracefully
• The browser uses the Web Audio API for real-time spectrum visualization — the server only sends fallback perception data
• Music directory can be changed live via the Library tab or via POST /api/set-music-dir
• Perception loop runs at 2fps server-side (idle when no clients connected)
• All POST bodies are limited to 64KB
• All rendered HTML is XSS-protected via escapeHtml()
• Graceful shutdown on SIGINT/SIGTERM