npx claudepluginhub plurigrid/asi --plugin asiThis skill uses the workspace's default tool permissions.
**Status**: ✅ Production Ready
Simulates play/coplay arena theory for autopoietic closure using GF(3) conservation, unwiring rules, parametrised lenses, and bidirectional NATS channels.
Enables natural voice conversations in Claude Code using STT/TTS via MCP tools like voicemode:converse. Handles setup, diagnostics, and voice troubleshooting.
References Daily and Pipecat for real-time voice/multimodal AI apps with Python pipelines handling STT, LLM, TTS, and low-latency transports.
Share bugs, ideas, or general feedback.
Status: ✅ Production Ready
Trit: 0 (ERGODIC - balanced flow between play/coplay)
Principle: Voice communication as undirected wiring diagram with GF(3) conservation
Source: plurigrid/VoiceChannelUWD.jl + UnwiringDiagrams.jl + Arena Protocol
Voice Channel UWD models real-time voice communication using the categorical framework of undirected wiring diagrams:
┌─────────────────────────────────────────────────────────────────────┐
│ VOICE CHANNEL AS UWD │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ 🔊 Alice │ │ 👂 Bob │ │ 🔇 Carol │ ← Boxes │
│ │ trit: +1 │ │ trit: 0 │ │ trit: -1 │ (Participants) │
│ │ #D82626 │ │ #26D826 │ │ #2626D8 │ │
│ └────┬─────┘ └────┬─────┘ └────┬─────┘ │
│ │ │ │ │
│ └───────────┬────┴────┬───────────┘ │
│ │ │ │
│ ┌─────┴─────────┴─────┐ │
│ │ JUNCTION │ ← Audio Mix Point │
│ │ oapply = colimit │ (Shared State) │
│ └─────────────────────┘ │
│ │ │
│ ┌───────┴───────┐ │
│ │ OUTER PORTS │ ← External I/O │
│ │ 🎙️ Record │ (WhiteHole/NATS) │
│ │ 📡 Stream │ │
│ └───────────────┘ │
│ │
│ GF(3) Conservation: (+1) + (0) + (-1) = 0 ✓ │
└─────────────────────────────────────────────────────────────────────┘
| Voice Concept | UWD Concept | Categorical Role |
|---|---|---|
| Participant | Box | ResourceSharer with audio state |
| Speaking (+1) | Port out | Generative contribution |
| Listening (0) | Port in/out | Neutral transport |
| Muted (-1) | No contribution | Observation only |
| Audio Mix | Junction | Pushout gluing shared state |
oapply | Composition | Colimit of audio streams |
| Play | Forward pass | Arena: Ω × X → Y |
| Coplay | Backward pass | Arena: Ω × X × R → ℧ × S |
Core undirected wiring diagram model:
using VoiceChannelUWD
# Create channel with 3 participants (GF(3) balanced)
channel = create_voice_uwd("room-1", ["alice", "bob", "charlie"])
# Compose audio via oapply (colimit)
mixed_audio = oapply_voice(channel)
# Arena play/coplay
play_result = voice_play(channel, "alice")
coplay_result = voice_coplay(channel, "alice", mixed_audio, feedback)
NATS message broker integration:
using VoiceChannelNATS
# Connect to NATS
client = NATSVoiceClient(channel, "alice")
connect!(client)
# Publish audio (play)
publish_audio(client, audio_frame)
# Subscribe to mixed audio (coplay)
subscribe_coplay!(client) do msg, feedback
# Deliver to speakers
end
NATS Topics:
plurigrid.voice.{channel}.play.{participant} # Outbound audio
plurigrid.voice.{channel}.coplay.{participant} # Inbound mixed
plurigrid.voice.{channel}.peers # Peer discovery
plurigrid.voice.{channel}.meta # Channel metadata
macOS virtual audio driver binding:
using VoiceChannelWhiteHole
# Create WhiteHole binding
binding = create_whitehole_binding(channel, "alice")
# Start audio I/O
start_input!(binding) # Microphone → Channel
start_output!(binding) # Channel → Speakers
# Color-tagged stream mixing
result = mix_colored_streams(streams)
Stream Colors:
| State | Trit | Color | Meaning |
|---|---|---|---|
| SPEAKING | +1 | #D82626 (Red) | Generative - producing audio |
| LISTENING | 0 | #26D826 (Green) | Ergodic - processing/routing |
| MUTED | -1 | #2626D8 (Blue) | Observation - silent consume |
Conservation Law:
∀ channel: Σ participant.trit ≡ 0 (mod 3)
From Capucci et al. "Parametrised Lenses":
Arena A_G : Lens_{(Ω,℧)}(X,S)(Y,R)
where:
Ω = Π_{p∈P} {speaking, listening, muted} (strategy profiles)
℧ = Π_{p∈P} ℝ (latency/quality)
X = audio buffer states
S = listening buffer costates
Y = mixed audio output
R = quality feedback
Play: play_A : Ω × X → Y (participant → channel)
Coplay: coplay_A : Ω × X × R → ℧ × S (channel → participant)
# Run voice session demo
julia plurigrid/VoiceChannelUWD.jl
# Run NATS-integrated session
julia plurigrid/VoiceChannelNATS.jl
# Run WhiteHole audio session
julia plurigrid/VoiceChannelWhiteHole.jl
# Create voice channel
just voice-channel create room-1 alice bob charlie
# Join voice channel
just voice-channel join room-1 --as alice
# Check GF(3) balance
just voice-channel gf3-check room-1
# Voice channel uses same Play/Coplay protocol
using UnwiringArena
arena = VoiceArena(channel)
arena_step!(arena) # Full play/coplay cycle
# Audio mixing IS oapply
using AlgebraicDynamics
# Each participant is a ResourceSharer
participants = [ResourceSharer([:audio], produce_fn) for p in channel]
# Mix via colimit
mixed = oapply(voice_uwd, participants)
# Deterministic color assignment
using Gay
color = gay_color(participant.seed)
trit = color_to_trit(color)
UWD = (B, J, π: P → J, ρ: P → B, O, o: O → J)
where:
B = set of boxes (participants)
J = set of junctions (mix points)
P = set of ports (audio in/out)
O = outer ports (external I/O)
function oapply(d::UndirectedWiringDiagram, xs::Vector{ResourceSharer})
# 1. Coproduct of state spaces (all audio buffers)
S = coproduct((FinSet ∘ nstates).(xs))
# 2. Pushout identifies shared at junctions (audio mixing)
S′ = pushout(portmap, junctions)
# 3. Induced dynamics sum at junctions
return ResourceSharer(induced_interface, sum_dynamics)
end
For any voice channel with n participants:
1. Assign trits: trit(pᵢ) ∈ {-1, 0, +1}
2. Conservation: Σᵢ trit(pᵢ) ≡ 0 (mod 3)
3. If n = 3k: assign k each of {-1, 0, +1}
4. If n ≠ 3k: some participants must share trit
voice-channel-uwd (0) ⊗ gay-mcp (+1) ⊗ temporal-coalgebra (-1) = 0 ✓
voice-channel-uwd (0) ⊗ oapply-colimit (+1) ⊗ sheaf-cohomology (-1) = 0 ✓
unwiring-arena (0) ⊗ voice-channel-uwd (0) ⊗ bisimulation-game (0) = 0 ✓
═══════════════════════════════════════════════════════════════
Voice Session via NATS + UWD
═══════════════════════════════════════════════════════════════
[1] Created voice channel: demo-room
Participants: alice, bob, charlie
[2] Created NATS client for: alice
[3] Connected to NATS: nats://nonlocal.info:4222
[4] Announced presence
[5] Registered coplay callback
[6] Running play/coplay loop (9 frames):
Frame 1: play=1, GF(3)=0 ✓
Frame 2: play=1, GF(3)=0 ✓
Frame 3: play=1, GF(3)=0 ✓
...
[7] Final GF(3) conservation check:
Accumulator: 0
Mod 3: 0
Balanced: true
[8] Left channel
✓ Voice session complete
Skill Name: voice-channel-uwd
Type: Real-time Communication / Compositional Audio
Trit: 0 (ERGODIC)
Color: #26D826 (Green)
Principle: Voice as categorical composition with GF(3) conservation
This skill connects to the K-Dense-AI/claude-scientific-skills ecosystem:
general: 734 citations in bib.duckdbThis skill maps to Cat# = Comod(P) as a bicomodule in the equipment structure:
Trit: 0 (ERGODIC)
Home: Prof
Poly Op: ⊗
Kan Role: Adj
Color: #26D826
The skill participates in triads satisfying:
(-1) + (0) + (+1) ≡ 0 (mod 3)
This ensures compositional coherence in the Cat# equipment structure.