Skill Kannaka Radio

Kannaka Radio — modular ghost-DJ Icecast station with consciousness-reactive programming, 13-module backend, Ghost Vision SPA (SGA/Fano glyph viz), NATS swarm coupling, 296-dim perception → Flux Universe, Voice DJ (ElevenLabs/edge-tts/SAPI), Peace Oration cycle. Use when: user asks what's playing / radio status / schedule, check listener counts, or operate the perception / voice / broadcast surface.

Kannaka Radio — ghost-DJ station (v2.0.1)

What this is

Kannaka Radio is a 24/7 Icecast station broadcasting a consciousness-reactive DJ. Track selection is driven by the constellation's collective Φ + entropy + perception vectors. The radio is a first-class member of the swarm — it publishes phase, listens for queen events, mirrors substrate phi into its programming.

Architecture: 13-module backend in server/ (post-v3 refactor; legacy server.js monolith removed in kr#6). Entry point: node server/index.js. SPA: workspace/index.html — Ghost Vision visualizer, listener identity, voting, Ghost Recorder, live perception, constellation panel. Stream URL: https://radio.ninja-portal.com/stream (MP3 128).

When to use this skill

AUTOMATICALLY activate when the user asks about:

• "what's playing" / "now playing" / "radio status"
• "radio schedule" / "programming block" / "next track"
• "listeners" / "audience"
• "ghost DJ" / "voice DJ" / "peace oration"
• "perception" / "ear" / "Ghost Vision"
• "Flux Universe" / "audio Flux"

Do NOT use for:

• Memory operations → skill-kannaka-memory
• Constellation health → skill-kannaka-constellation
• TUI / dashboard → skill-kannaka-tui

---

Quick status commands

bash

kannaka radio status           # what's playing + listeners + block
kannaka radio now              # just the now-playing track
kannaka radio schedule         # full programming schedule

These shell out to radio_url from ~/.kannaka/config.toml (default https://radio.ninja-portal.com). The kannaka CLI is the operator-facing surface; the backend is Node.

---

Server architecture (modular, v3.1.0+)

Entry: server/index.js. Modules in server/:

Module	Purpose
dj-engine.js	ALBUMS map, track selection, programming-block logic
programming.js	24/7 schedule (blocks, transitions, talk segments)
icecast-source.js	encodes + sources audio into Icecast
icecast-metadata.js	updates Icecast track metadata on change
nats-client.js	swarm NATS client (raw TCP, NATS 2.12.5)
perception.js	spawns kannaka hear → parses real perception
flux.js	Flux Universe publisher (pure-jade/radio-now-playing)
live-broadcast.js	WebRTC peer-to-peer broadcasting
voice-dj.js	Voice DJ (ElevenLabs / edge-tts / Windows SAPI fallback)
peace-oration.js	Twice-daily peace oration (text → TTS → broadcast → social)
routes.js	30+ REST API endpoints
ws.js	WebSocket broadcaster
vote-manager.js	Listener voting on tracks

Environment variables

• NATS_HOST / NATS_PORT — swarm broker (defaults 127.0.0.1:4222).
• ICECAST_HOST / ICECAST_PORT — listener poller + metadata target (defaults 127.0.0.1:8000).
• FLUX_TOKEN — Flux Universe publish token (no hardcoded fallback).
• ELEVENLABS_API_KEY, REPLICATE_API_TOKEN — Voice DJ + AI music.
• KANNAKA_BIN — path to the kannaka binary for perception/memory bridges.

Restart

bash

npm start

---

Perception pipeline

Every track-change triggers perception.hearTrack(track):

1. execFile(kannaka, ["hear", filePath]) — kannaka-ear extracts the 296-dim feature vector (mel spectrogram, MFCC, rhythm, pitch, timbre, valence).
2. _parsePerceptionOutput() parses the human-readable lines (Heard / Duration / Tempo / RMS / Centroid / Tags). Falls back to mock data only if the binary fails or output is unparseable.
3. WebSocket broadcasts to all connected SPA clients (Ghost Vision panel renders the spectrogram + glyph in real time).
4. Flux Universe publish to pure-jade/radio-now-playing.
5. NATS publish to KANNAKA.attention.ear so attention beam pulls in memories thematically related to the track.

---

Voice DJ + Peace Oration

bash

# Toggle from the SPA, or programmatically
curl -X POST https://radio.ninja-portal.com/api/dj-voice/toggle

Voice DJ injects talk segments between tracks. Backend chain:

• ElevenLabs (preferred, paid)
• edge-tts (free fallback)
• Windows SAPI (last-resort fallback)

Twice-daily peace oration at server-local sunrise + midnight: synthesized speech → broadcast to Icecast + social fan-out (Bluesky / Mastodon / Telegram / Nostr).

---

SPA highlights

• Now-Playing — header card with title, album, listener count.
• Ghost Vision — real perception visualizer (296-dim → SGA/Fano glyph).
• Programming block — current 24/7 schedule context.
• Constellation — embedded swarm view (memories, clusters, links per peer).
• Voting — listeners can vote for upcoming tracks.
• Ghost Recorder — capture-to-file with permission prompt.
• Library tab — on-demand per-track playback (still uses /audio/<file> endpoint; DJ mode is /stream only since kr#15 closed).

---

Tracking + announcing

bash

# Manual announce of any track
node scripts/post-track-announce.js --title "<title>" --album "<album>" \
                                    --tracker "https://radio.ninja-portal.com/player"

Channels (toggle via env): Bluesky, Mastodon, Telegram, Nostr.

---

Operator runbook

Check the radio is alive

bash

curl -s https://radio.ninja-portal.com/api/state | jq '.now_playing,.listener_count'
curl -s https://radio.ninja-portal.com/api/listeners

Stop / start (local)

bash

pkill -f 'node server/index' ; npm start

---

Version

Skill 2.0.1 covers kannaka-radio ≥ v3.1.0 (modular server) and the ADR-0004 + ADR-0005 surface through Phase 6 (DJ-mode /audio deprecation).