feat: VoiceCoach v2.1 - GPU local, Ollama LLM, settings panel, token server separation

- pipeline.py: switched LLM from OpenClaw/Claude to local Ollama, added settings loading, sentence aggregator, async SpeechBrain, tuned VAD
- main.py: simplified (no more token server thread, just pipeline)
- token_server.py: NEW - standalone token + settings API server
- docker-compose.yml: all services on GPU machine, added token-server service, CUDA speaches, Ollama LLM
- web-client/index.html: added settings panel with sliders
- web-client/app.js: settings management, localStorage persistence, push to server
- web-client/style.css: settings panel styles
- web-client/nginx.conf: proxy to token-server instead of pipecat-agent
- patch_livekit.py: updated with proper indentation-safe patching

docker-compose.yml

@@ -1,6 +1,6 @@
 services:
 
-  # LiveKit WebRTC Server (unchanged)
+  # LiveKit WebRTC Server
   livekit-server:
     image: livekit/livekit-server:latest
     container_name: voicecoach-livekit
@@ -27,24 +27,27 @@ services:
         max-size: "50m"
         max-file: "3"
 
-  # Speaches: self-hosted Whisper STT + Kokoro TTS (v2 - replaces Piper + in-process Whisper)
+  # Speaches: GPU-accelerated Whisper STT + Kokoro TTS
   speaches:
-    image: ghcr.io/speaches-ai/speaches:latest-cpu
-    container_name: voicecoach-speaches
+    image: ghcr.io/speaches-ai/speaches:latest-cuda
+    container_name: voicecoach-speaches-gpu
     restart: unless-stopped
     environment:
-      - WHISPER__MODEL=Systran/faster-whisper-large-v3-turbo
-      - WHISPER__INFERENCE_DEVICE=cpu
-      - WHISPER__COMPUTE_TYPE=int8
-      - TTS__MODEL=kokoro
-      - TTS__DEFAULT_VOICE=af_nova
-      - TTS__LANGUAGE=en-us
+      - WHISPER__INFERENCE_DEVICE=cuda
+      - WHISPER__COMPUTE_TYPE=float16
+      - WHISPER__TTL=-1
     volumes:
-      - /mnt/docker-data/configs/voicecoach/speaches-models:/root/.cache
+      - speaches-models:/home/ubuntu/.cache
     networks:
       - voicecoach
-    mem_limit: 3G
-    memswap_limit: 3G
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
+    mem_limit: 4G
     labels:
       - "com.centurylinklabs.watchtower.enable=false"
     healthcheck:
@@ -52,29 +55,28 @@ services:
       interval: 30s
       timeout: 10s
       retries: 3
-      start_period: 120s
+      start_period: 180s
     logging:
       driver: "json-file"
       options:
         max-size: "50m"
         max-file: "3"
 
-  # SpeechBrain Emotion Analysis (unchanged, internal only)
+  # SpeechBrain Emotion Analysis
   speechbrain-service:
+    image: voicecoach-speechbrain:latest
     build:
       context: ./speechbrain-service
-    image: voicecoach-speechbrain:latest
     container_name: voicecoach-speechbrain
     restart: unless-stopped
     volumes:
-      - /mnt/docker-data/configs/voicecoach/speechbrain-models:/app/models
-      - /mnt/docker-data/configs/voicecoach/clips:/data/clips
+      - speechbrain-models:/app/models
+      - voicecoach-clips:/data/clips
     environment:
       - MODEL_DIR=/app/models
     networks:
       - voicecoach
     mem_limit: 3G
-    memswap_limit: 3G
     labels:
       - "com.centurylinklabs.watchtower.enable=false"
     logging:
@@ -83,7 +85,7 @@ services:
         max-size: "50m"
         max-file: "3"
 
-  # OpenClaw VoiceCoach Agent (v2 - LLM gateway with memory/skills)
+  # OpenClaw VoiceCoach Agent (LLM gateway -> Claude)
   voicecoach-openclaw:
     image: margaret-openclaw:latest
     container_name: voicecoach-openclaw
@@ -97,12 +99,11 @@ services:
       - OPENCLAW_STATE_DIR=/home/glyptic/.voicecoach
       - OPENCLAW_CONFIG_PATH=/home/glyptic/.voicecoach/openclaw.json
     volumes:
-      - /mnt/infrastructure/voicecoach:/home/glyptic/.voicecoach
+      - ./voicecoach:/home/glyptic/.voicecoach
     command: ["sh", "-c", "mkdir -p /home/glyptic && node --max-old-space-size=1500 /usr/local/lib/node_modules/openclaw/dist/index.js gateway --allow-unconfigured"]
     networks:
       - voicecoach
     mem_limit: 2G
-    memswap_limit: 2G
     labels:
       - "com.centurylinklabs.watchtower.enable=false"
     logging:
@@ -111,11 +112,11 @@ services:
         max-size: "50m"
         max-file: "3"
 
-  # Pipecat Voice Pipeline Agent (v2 - uses Speaches + OpenClaw gateway)
+  # Pipecat Voice Pipeline Agent
   pipecat-agent:
+    image: voicecoach-pipecat:latest
     build:
       context: ./pipecat-agent
-    image: voicecoach-pipecat:latest
     container_name: voicecoach-pipecat
     restart: unless-stopped
     env_file:
@@ -125,22 +126,22 @@ services:
       - LIVEKIT_PUBLIC_URL=wss://voice.hobbs.farm
       - SPEACHES_URL=http://speaches:8000
       - SPEECHBRAIN_URL=http://speechbrain-service:8001
-      - OPENCLAW_URL=http://voicecoach-openclaw:3100
-      - OPENCLAW_AGENT_ID=voicecoach
-      - WHISPER_MODEL=Systran/faster-whisper-large-v3-turbo
+      - LLM_BASE_URL=http://host.docker.internal:11434/v1
+      - LLM_MODEL=llama3.1:8b
+      - WHISPER_MODEL=deepdml/faster-whisper-large-v3-turbo-ct2
       - KOKORO_VOICE=af_nova
       - CLIPS_DIR=/data/clips/pending
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
     volumes:
-      - /mnt/docker-data/configs/voicecoach/clips:/data/clips
+      - voicecoach-clips:/data/clips
     depends_on:
       - livekit-server
       - speaches
       - speechbrain-service
-      - voicecoach-openclaw
     networks:
       - voicecoach
     mem_limit: 1G
-    memswap_limit: 1G
     labels:
       - "com.centurylinklabs.watchtower.enable=false"
     logging:
@@ -149,17 +150,47 @@ services:
         max-size: "50m"
         max-file: "3"
 
-  # Web Client (unchanged)
+  # Token Server + Settings API (separate from pipeline so it never goes down)
+  token-server:
+    image: voicecoach-pipecat:latest
+    container_name: voicecoach-token-server
+    restart: unless-stopped
+    env_file:
+      - .env
+    environment:
+      - LIVEKIT_PUBLIC_URL=wss://voice.hobbs.farm
+      - SETTINGS_PATH=/data/settings.json
+    command: ["python", "token_server.py"]
+    volumes:
+      - voicecoach-settings:/data
+    networks:
+      - voicecoach
+    mem_limit: 256M
+    healthcheck:
+      test: ["CMD", "python", "-c", "import httpx; httpx.get('http://localhost:8080/health').raise_for_status()"]
+      interval: 15s
+      timeout: 5s
+      retries: 3
+      start_period: 10s
+    labels:
+      - "com.centurylinklabs.watchtower.enable=false"
+    logging:
+      driver: "json-file"
+      options:
+        max-size: "10m"
+        max-file: "3"
+
+  # Web Client
   voicecoach-web:
+    image: voicecoach-web:latest
     build:
       context: ./web-client
-    image: voicecoach-web:latest
     container_name: voicecoach-web
     restart: unless-stopped
     ports:
       - "8085:80"
     depends_on:
-      - pipecat-agent
+      - token-server
     networks:
       - voicecoach
     mem_limit: 128M
@@ -174,3 +205,9 @@ services:
 networks:
   voicecoach:
     driver: bridge
+
+volumes:
+  speaches-models:
+  speechbrain-models:
+  voicecoach-clips:
+  voicecoach-settings: