feat: Tarife einstellbar + 15 weitere Filler + Mitteilung bei Leer-Antwort

- Strompreis (EUR/kWh), Notruf-Kosten (EUR/Alarm), Cartesia ($/Zeichen) jetzt im
  Admin live einstellbar (RUNTIME_SETTABLE).
- OPENING-Beruhigungssaetze 12 -> 27 (random.choice; formale "Sie"-Uebersetzung
  ersetzt beim Neustart den alten "dich"-Cache).
- Leer-/kein-Treffer-Antwort: statt Stille eine freundliche lokalisierte
  Mitteilung ("Dazu konnte ich keine Informationen finden."), gesprochen.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

app/core/orchestrator.py

@@ -10,6 +10,21 @@ from app.metrics import timer, metrics
 # zwischen Geduldssätzen, bei noch laufender Recherche.
 FILLER_PATIENCE_INTERVAL = 10.0
 
+# Freundliche Mitteilung statt Stille, wenn keine Antwort zustande kommt
+# (z. B. Web-Suche ohne Treffer). Senior wartet sonst vergeblich. Fallback Deutsch.
+_NO_RESULT = {
+    "de": "Dazu konnte ich leider keine Informationen finden.",
+    "en": "I'm sorry, I couldn't find any information about that.",
+    "nl": "Daarover kon ik helaas geen informatie vinden.",
+    "fr": "Je suis désolé, je n'ai trouvé aucune information à ce sujet.",
+    "es": "Lo siento, no he encontrado información al respecto.",
+    "it": "Mi dispiace, non ho trovato informazioni in merito.",
+}
+
+
+def _no_result_message(language: str | None) -> str:
+    return _NO_RESULT.get((language or "de").lower(), _NO_RESULT["de"])
+
 
 def _stage(name: str):
     return timer("stage_duration_seconds", {"stage": name})
@@ -135,8 +150,8 @@ class Orchestrator:
                 history=history,
                 language=effective_language,
             )
-        if not trace.semantic_response:
-            raise RuntimeError("LLM returned an empty response")
+        if not (trace.semantic_response or "").strip():
+            trace.semantic_response = _no_result_message(effective_language)
 
         trace.spoken_response = await self.spoken_adapter.run(
             trace.semantic_response,
@@ -330,9 +345,16 @@ class Orchestrator:
                     for sentence in chunker.feed(result):
                         await _dispatch(sentence)
 
-            trace.semantic_response = "".join(parts)
+            trace.semantic_response = "".join(parts).strip()
             if not trace.semantic_response:
-                raise RuntimeError("LLM returned an empty response")
+                # Keine Antwort (z. B. Suche ohne Treffer) -> freundliche Mitteilung
+                # statt Stille, damit der Nutzer nicht vergeblich wartet.
+                trace.semantic_response = _no_result_message(effective_language)
+                if on_token:
+                    await on_token(trace.semantic_response)
+                if chunker:
+                    for sentence in chunker.feed(trace.semantic_response):
+                        await _dispatch(sentence)
             trace.citations = collected_citations
             trace.web_searches = search_count[0]
 

app/pipeline/fillers.py

@@ -35,6 +35,21 @@ OPENING_PHRASES = [
     "Give me a second, I'll find that out for you.",
     "I'm checking that for you now, one moment please.",
     "Let me look that up so I can tell you exactly.",
+    "Let me check the current details for you — it won't take long.",
+    "I'll find the up-to-date answer for you, one moment.",
+    "Allow me a brief moment to look into that.",
+    "Let me gather the latest information for you.",
+    "I'm on it — just give me a short moment.",
+    "Let me make sure I have the current facts for you.",
+    "One moment please, I'm fetching the latest for you.",
+    "Let me quickly check what's current on that.",
+    "I'll look into that for you right away, just a moment.",
+    "Bear with me a second while I find that out.",
+    "Let me get you the most recent information on that.",
+    "Just a brief moment — I'm checking the latest for you.",
+    "I'll dig that up for you now, one moment please.",
+    "Let me find the current answer so I can tell you properly.",
+    "Let me take a quick look and find that out for you.",
 ]
 
 PATIENCE_PHRASES = [

app/runtime_config.py

@@ -35,6 +35,9 @@ RUNTIME_SETTABLE: dict[str, tuple[str, str, str]] = {
     "memory_extraction_every_n_turns": ("Extraktion alle N Turns",     "int",   "z.B. 3"),
     "daily_request_limit":           ("Tageskontingent (global)",      "int",   "0 = unbegrenzt"),
     "cost_daily_limit_usd":          ("Kostenbudget/Tag (global, $)",  "float", "0 = unbegrenzt — pro Nutzer überschreibbar"),
+    "electricity_price_eur_per_kwh": ("Strompreis (€/kWh)",            "float", "Kosten lokaler LLM-Anfragen (z.B. 0.33)"),
+    "emergency_alert_eur":           ("Notruf-Kosten (€/Alarm)",       "float", "SMS+Anruf pro Alarm (z.B. 0.20)"),
+    "cartesia_usd_per_char":         ("Cartesia-TTS ($/Zeichen)",      "float", "z.B. 0.00005 (100k Credits ≈ 100.000 Zeichen)"),
     "emergency_cooldown_minutes":    ("Notruf-Sperre (Minuten)",       "int",   "Erneuter Alarm für N Minuten blockiert (Standard 5, 0 = aus)"),
 }