refactor: VAD upgrade, retry, dead code cleanup, AGENT removal

- audio: switch VAD to webrtcvad with RMS gate + fallback to RMS
- audio: honor FOLLOWUP_TIMEOUT — short silence wait after bot response
- llm: retry with exponential backoff on network errors and 5xx
- llm: VOICE_MAX_TOKENS env (default 300) instead of hardcoded 150
- tts: optional VAD-based barge-in (BARGE_IN_ENABLED, off by default)
- tts: remove dead start_barge_in_listener / was_barge_in helpers
- config: drop AGENT/LUSYA_AGENT — routing happens via session_key
- modes: remove unused imports, pass FOLLOWUP_TIMEOUT to follow-up record()
- docs: full rewrite of README and CLAUDE.md to match current architecture

satellite/config.py

@@ -20,15 +20,14 @@ logging.basicConfig(
 log = logging.getLogger("cosmo")
 
 # OpenClaw Gateway — Cosmo (по умолчанию)
+# Роутинг к нужному агенту делается через x-openclaw-session-key, поэтому AGENT не нужен.
 GATEWAY_URL   = os.getenv("GATEWAY_URL", "http://192.168.31.103:18789")
 GATEWAY_TOKEN = os.getenv("GATEWAY_TOKEN")
-AGENT         = os.getenv("AGENT", "openclaw/main")
 VOICE_MODEL   = os.getenv("VOICE_MODEL", "openai/gpt-4o-mini")
 
 # OpenClaw Gateway — Люся
 LUSYA_GATEWAY_URL   = os.getenv("LUSYA_GATEWAY_URL", "http://192.168.31.103:18790")
 LUSYA_GATEWAY_TOKEN = os.getenv("LUSYA_GATEWAY_TOKEN", GATEWAY_TOKEN)
-LUSYA_AGENT         = os.getenv("LUSYA_AGENT", "openclaw/wife")
 LUSYA_VOICE_MODEL   = os.getenv("LUSYA_VOICE_MODEL", VOICE_MODEL)
 
 # Keep-alive HTTP сессии — переиспользуют TCP/TLS соединения
@@ -46,20 +45,16 @@ AGENTS = {
     "cosmo": {
         "name": "Cosmo",
         "gateway_url": GATEWAY_URL,
-        "token": GATEWAY_TOKEN,
-        "agent": AGENT,
         "voice_model": VOICE_MODEL,
-        "session_key": os.getenv("COSMO_SESSION_KEY", "voice:home:cosmo"),
+        "session_key": os.getenv("COSMO_SESSION_KEY", "agent:main:voice:home"),
         "tts_voice": os.getenv("COSMO_TTS_VOICE", ""),
         "session": _make_session(GATEWAY_TOKEN),
     },
     "lusya": {
         "name": "Люся",
         "gateway_url": LUSYA_GATEWAY_URL,
-        "token": LUSYA_GATEWAY_TOKEN,
-        "agent": LUSYA_AGENT,
         "voice_model": LUSYA_VOICE_MODEL,
-        "session_key": os.getenv("LUSYA_SESSION_KEY", "voice:home:lusya"),
+        "session_key": os.getenv("LUSYA_SESSION_KEY", "agent:wife:voice:home"),
         "tts_voice": os.getenv("LUSYA_TTS_VOICE", ""),
         "session": _make_session(LUSYA_GATEWAY_TOKEN),
     },
@@ -73,6 +68,18 @@ SILENCE_THRESHOLD = int(os.getenv("SILENCE_THRESHOLD", "500"))
 SILENCE_DURATION  = float(os.getenv("SILENCE_DURATION", "1.5"))
 MAX_DURATION      = int(os.getenv("MAX_DURATION", "15"))
 FOLLOWUP_TIMEOUT  = float(os.getenv("FOLLOWUP_TIMEOUT", "8"))
+VAD_AGGRESSIVENESS = int(os.getenv("VAD_AGGRESSIVENESS", "2"))  # webrtcvad 0..3
+
+# LLM
+VOICE_MAX_TOKENS  = int(os.getenv("VOICE_MAX_TOKENS", "300"))
+LLM_RETRIES       = int(os.getenv("LLM_RETRIES", "3"))
+
+# Barge-in (прерывание TTS голосом)
+# Работает только при разнесённых колонке/мике или в наушниках — иначе эхо собственного TTS
+# будет триггерить прерывание. По умолчанию выключен.
+BARGE_IN_ENABLED   = os.getenv("BARGE_IN_ENABLED", "false").lower() in ("1", "true", "yes")
+BARGE_IN_THRESHOLD = int(os.getenv("BARGE_IN_THRESHOLD", "1500"))  # RMS, обычно > SILENCE_THRESHOLD
+BARGE_IN_WARMUP    = float(os.getenv("BARGE_IN_WARMUP", "0.8"))    # сек пропуска в начале TTS
 
 # Groq client
 groq_client = Groq(api_key=os.getenv("GROQ_API_KEY"))