Feat: Hermes Voice Client — CustomTkinter GUI + Push-to-Talk + .exe Build

- main.py: CustomTkinter GUI (400x520) mit Status-Indikator, Conversation-Log,
  Screenshot-Toggle und Settings-Dialog. Thread-sicherer UI-Queue für Hotkey-Callbacks.
- hotkey_listener.py: pynput globaler Hotkey (press/release) + KeyCapturer für
  interaktive Hotkey-Konfiguration im Settings-Dialog.
- config_manager.py: JSON-Settings in ~/.hermes-voice/settings.json mit DEFAULTS,
  load() merged gespeicherte mit Default-Werten.
- audio_input.py: Vereinfacht auf start_recording/stop_recording/transcribe (kein
  Wake-Word-Loop mehr, direkt hotkey-gesteuert).
- ai_client.py: Settings-dict statt config.py-Imports, MC2-URL/Modelle konfigurierbar.
- requirements.txt: customtkinter>=5.2.0 + pynput>=1.7.6 hinzugefügt, pystray entfernt.
- build.bat: PyInstaller --onefile --windowed → dist/HermesVoice.exe.
- setup.bat: Veraltete Wake-Word-Hinweise entfernt, GUI-Check angepasst.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

client/hermes-voice/ai_client.py

@@ -1,15 +1,20 @@
 import httpx
-from config import (
-    MC2_BASE_URL, CHAT_MODEL, VISION_MODEL,
-    MAX_RESPONSE_TOKENS, REQUEST_TIMEOUT, SYSTEM_PROMPT,
-    SCREENSHOT_ON_QUERY,
+
+SYSTEM_PROMPT = (
+    "Du bist Hermes, ein hilfreicher KI-Assistent. "
+    "Antworte kurz und präzise, da deine Antwort vorgelesen wird. "
+    "Maximal 3-4 Sätze, kein Markdown."
 )
+REQUEST_TIMEOUT = 30
 
 
-def ask(text: str, screenshot_b64: str | None = None) -> str:
+def ask(text: str, screenshot_b64: str | None, settings: dict) -> str:
     """Schickt Text (+ optionalen Screenshot) an MC2 und gibt die Antwort zurück."""
-    use_vision = screenshot_b64 is not None and SCREENSHOT_ON_QUERY
-    model = VISION_MODEL if use_vision else CHAT_MODEL
+    use_vision = screenshot_b64 is not None
+    model = settings.get("vision_model", "Qwen3-VL-2B-Instruct") if use_vision \
+        else settings.get("chat_model", "Hermes-4-14B")
+    base_url = settings.get("mc2_url", "http://192.168.178.151:9001/v1")
+    max_tokens = int(settings.get("max_response_tokens", 200))
 
     if use_vision:
         user_content = [
@@ -27,16 +32,13 @@ def ask(text: str, screenshot_b64: str | None = None) -> str:
             {"role": "system", "content": SYSTEM_PROMPT},
             {"role": "user", "content": user_content},
         ],
-        "max_tokens": MAX_RESPONSE_TOKENS,
+        "max_tokens": max_tokens,
         "stream": False,
     }
 
     try:
         with httpx.Client(timeout=REQUEST_TIMEOUT) as client:
-            response = client.post(
-                f"{MC2_BASE_URL}/chat/completions",
-                json=payload,
-            )
+            response = client.post(f"{base_url}/chat/completions", json=payload)
             response.raise_for_status()
             return response.json()["choices"][0]["message"]["content"].strip()
     except httpx.TimeoutException: