From 7d30fe9e7c818978ac9803b7077ec79e7d53d774 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Dieter=20Schl=C3=BCter?= <dieter.schlueter@linix.de>
Date: Sun, 21 Jun 2026 01:10:51 +0200
Subject: [PATCH 01/10] =?UTF-8?q?ui(tts):=20Ton-Presets=20=E2=80=94=20Schn?=
 =?UTF-8?q?ell=20(piper),=20Hohe=20Qualit=C3=A4t=20(chatterbox),=20Cloud?=
 =?UTF-8?q?=20(openrouter)=20zur=C3=BCck?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Vorlesen-Presets wieder nach Tempo/Qualität benannt: 📱 Im Gerät · ⚡ Schnell ·
  ✨ Hohe Qualität · ☁ Cloud. Cloud (OpenRouter) wieder als sichtbarer Knopf.
- syncPresetUI: exakte Markierung (openrouter hat eigenen Knopf, kein Sonderfall mehr).
- Doku §5.1.2 angepasst.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 BEDIENUNGSANLEITUNG.md |  2 +-
 app/web/app.js         |  3 +--
 app/web/index.html     | 19 +++++++++++++------
 3 files changed, 15 insertions(+), 9 deletions(-)

diff --git a/BEDIENUNGSANLEITUNG.md b/BEDIENUNGSANLEITUNG.md
index 1c5be23..de4ae5b 100644
--- a/BEDIENUNGSANLEITUNG.md
+++ b/BEDIENUNGSANLEITUNG.md
@@ -800,7 +800,7 @@ https://va.beispiel.de/        ← remote über Reverse-Proxy (alles, inkl. Mikr
 
 | Element | Funktion |
 |---------|----------|
-| **Vorlesen** | Wie wird die Antwort vorgelesen? `📱 Im Gerät` = das Handy liest selbst vor (kein Server-Audio → spart Daten) · `☁ Server` = piper (lokal, zuverlässig) · `✨ Beste Qualität` = chatterbox (natürliche Stimme). „Im Gerät" erscheint nur, wo der Browser es unterstützt. |
+| **Vorlesen** | Wie wird die Antwort vorgelesen? `📱 Im Gerät` = das Handy liest selbst vor (kein Server-Audio → spart Daten; fällt auf Server zurück, wenn der Browser keine Stimmen hat) · `⚡ Schnell` = piper (lokal) · `✨ Hohe Qualität` = chatterbox (natürliche Stimme) · `☁ Cloud` = OpenRouter (Internet nötig). „Im Gerät" erscheint nur, wo der Browser es unterstützt. |
 | **✎ Neues Gespräch** | Frische Sitzung (Verlauf zurücksetzen) |
 | **🌙 Tag-/Nachtmodus** | Heller/dunkler Modus; folgt sonst dem Betriebssystem |
 | **Konto** | „Angemeldet als …", **⚙ Admin** (nur Admins → § 7.5), **Abmelden** |
diff --git a/app/web/app.js b/app/web/app.js
index 95c7cd3..d3d6ae5 100644
--- a/app/web/app.js
+++ b/app/web/app.js
@@ -164,8 +164,7 @@ function applyPlaybackChoice(serverProvider) {
 function syncPresetUI() {
   const val = ttsSel ? ttsSel.value : "";
   document.querySelectorAll(".tts-preset").forEach((btn) => {
-    const active = btn.dataset.tts === val ||
-      (val === "openrouter" && btn.dataset.tts === "chatterbox");  // Cloud -> "Beste Qualität"
+    const active = btn.dataset.tts === val;
     btn.classList.toggle("ring-2", active);
     btn.classList.toggle("ring-blue-500", active);
     btn.classList.toggle("border-blue-500", active);
diff --git a/app/web/index.html b/app/web/index.html
index 002d30f..7750b84 100644
--- a/app/web/index.html
+++ b/app/web/index.html
@@ -260,16 +260,23 @@
         </button>
         <button type="button" data-tts="piper"
           class="tts-preset w-full flex items-center gap-3 rounded-lg border border-slate-200 dark:border-slate-700 px-3 py-2 text-left hover:bg-slate-50 dark:hover:bg-slate-700/50 transition-colors">
-          <span class="text-lg shrink-0">☁</span>
-          <span class="flex-1 min-w-0"><span class="block text-sm font-medium">Server</span>
-            <span class="block text-xs text-slate-500 dark:text-slate-400">Standard · zuverlässig</span></span>
+          <span class="text-lg shrink-0">⚡</span>
+          <span class="flex-1 min-w-0"><span class="block text-sm font-medium">Schnell</span>
+            <span class="block text-xs text-slate-500 dark:text-slate-400">Piper · lokal, zuverlässig</span></span>
           <span class="check text-blue-600 dark:text-blue-400 font-bold opacity-0">✓</span>
         </button>
         <button type="button" data-tts="chatterbox"
           class="tts-preset w-full flex items-center gap-3 rounded-lg border border-slate-200 dark:border-slate-700 px-3 py-2 text-left hover:bg-slate-50 dark:hover:bg-slate-700/50 transition-colors">
           <span class="text-lg shrink-0">✨</span>
-          <span class="flex-1 min-w-0"><span class="block text-sm font-medium">Beste Qualität</span>
-            <span class="block text-xs text-slate-500 dark:text-slate-400">Natürliche Stimme</span></span>
+          <span class="flex-1 min-w-0"><span class="block text-sm font-medium">Hohe Qualität</span>
+            <span class="block text-xs text-slate-500 dark:text-slate-400">Chatterbox · natürliche Stimme</span></span>
+          <span class="check text-blue-600 dark:text-blue-400 font-bold opacity-0">✓</span>
+        </button>
+        <button type="button" data-tts="openrouter"
+          class="tts-preset w-full flex items-center gap-3 rounded-lg border border-slate-200 dark:border-slate-700 px-3 py-2 text-left hover:bg-slate-50 dark:hover:bg-slate-700/50 transition-colors">
+          <span class="text-lg shrink-0">☁</span>
+          <span class="flex-1 min-w-0"><span class="block text-sm font-medium">Cloud</span>
+            <span class="block text-xs text-slate-500 dark:text-slate-400">OpenRouter · benötigt Internet</span></span>
           <span class="check text-blue-600 dark:text-blue-400 font-bold opacity-0">✓</span>
         </button>
       </div>
@@ -318,6 +325,6 @@
     <div id="status" class="w-full max-w-3xl mx-auto mt-1 min-h-[1rem] text-xs text-slate-500 dark:text-slate-400"></div>
   </footer>
 
-  <script src="/app.js?v=39"></script>
+  <script src="/app.js?v=40"></script>
 </body>
 </html>

From 53ec2d7ec701ec026269dc897c007ecb65a2967a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Dieter=20Schl=C3=BCter?= <dieter.schlueter@linix.de>
Date: Sun, 21 Jun 2026 01:28:21 +0200
Subject: [PATCH 02/10] =?UTF-8?q?ui:=20"Angemeldet=20als=20=E2=80=A6"=20+?=
 =?UTF-8?q?=20"Abmelden"=20zur=C3=BCck=20in=20eine=20Leiste=20unter=20die?=
 =?UTF-8?q?=20Kopfzeile?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Identitätsleiste (Nutzer links, Abmelden-Link rechts) wieder direkt unter dem Header
  statt im ⋮-Menü. Logout-Link kommt wie bisher aus SSO_LOGOUT_URL.
- Admin-Button bleibt im ⋮-Menü (als "⚙ Admin-Bereich"); IDs unverändert.
- Doku §5.1.2 angepasst.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 BEDIENUNGSANLEITUNG.md | 19 +++++++++++--------
 app/web/index.html     | 23 ++++++++++++-----------
 2 files changed, 23 insertions(+), 19 deletions(-)

diff --git a/BEDIENUNGSANLEITUNG.md b/BEDIENUNGSANLEITUNG.md
index de4ae5b..f03cbb0 100644
--- a/BEDIENUNGSANLEITUNG.md
+++ b/BEDIENUNGSANLEITUNG.md
@@ -776,16 +776,16 @@ https://va.beispiel.de/        ← remote über Reverse-Proxy (alles, inkl. Mikr
 ┌─────────────────────────────────────────────────────────┐
 │  👄 Voice Assistant ............... [🇩🇪 ▾]   [ ⋮ ]      │
 ├─────────────────────────────────────────────────────────┤      ┌─ Menü (⋮) ──────────────┐
-│                                                         │      │ VORLESEN                 │
-│   Nachrichtenverlauf                                    │      │  ▣ 📱 Im Gerät           │
-│   (eigene Nachrichten: blaue Blase rechts)              │      │  ▢ ☁ Server              │
-│   (Assistent: graue Blase links)                        │      │  ▢ ✨ Beste Qualität     │
+│  Angemeldet als <user> ....................... Abmelden │      │ VORLESEN                 │
+├─────────────────────────────────────────────────────────┤      │  ▣ 📱 Im Gerät           │
+│   Nachrichtenverlauf                                    │      │  ▢ ⚡ Schnell            │
+│   (eigene Nachrichten: blaue Blase rechts)              │      │  ▢ ✨ Hohe Qualität      │
+│   (Assistent: graue Blase links)                        │      │  ▢ ☁ Cloud               │
 │                                                         │      │ ──────────────────────  │
 ├─────────────────────────────┬───────────────────────────┤      │  ✎ Neues Gespräch        │
 │  Texteingabe …   [Senden]   │  [🎤]                     │      │  🌙 Tag-/Nachtmodus      │
-└─────────────────────────────┴───────────────────────────┘      │ ──────────────────────  │
-  Statuszeile: „denkt …" / „verarbeite Sprache …" / leer          │  Angemeldet · Admin · Ab │
-                                                                  └──────────────────────────┘
+└─────────────────────────────┴───────────────────────────┘      │  ⚙ Admin-Bereich         │
+  Statuszeile: „denkt …" / „verarbeite Sprache …" / leer          └──────────────────────────┘
 ```
 
 **Kopfzeile (immer schlank):**
@@ -796,6 +796,9 @@ https://va.beispiel.de/        ← remote über Reverse-Proxy (alles, inkl. Mikr
 | **Sprache ▾** | Antwortsprache (→ § 6.6): `🔄 Flex` = folgt der gesprochenen Sprache · feste Sprache (🇩🇪/🇬🇧/…) = Antwort + Stimme in dieser Sprache |
 | **⋮ Menü** | Öffnet die Einstellungen (unten) |
 
+**Identitätsleiste (direkt unter der Kopfzeile):** links „Angemeldet als …" (SSO-Identität;
+„Gast" ohne SSO), rechts **Abmelden** (Link zum SSO-Logout).
+
 **Im ⋮-Menü:**
 
 | Element | Funktion |
@@ -803,7 +806,7 @@ https://va.beispiel.de/        ← remote über Reverse-Proxy (alles, inkl. Mikr
 | **Vorlesen** | Wie wird die Antwort vorgelesen? `📱 Im Gerät` = das Handy liest selbst vor (kein Server-Audio → spart Daten; fällt auf Server zurück, wenn der Browser keine Stimmen hat) · `⚡ Schnell` = piper (lokal) · `✨ Hohe Qualität` = chatterbox (natürliche Stimme) · `☁ Cloud` = OpenRouter (Internet nötig). „Im Gerät" erscheint nur, wo der Browser es unterstützt. |
 | **✎ Neues Gespräch** | Frische Sitzung (Verlauf zurücksetzen) |
 | **🌙 Tag-/Nachtmodus** | Heller/dunkler Modus; folgt sonst dem Betriebssystem |
-| **Konto** | „Angemeldet als …", **⚙ Admin** (nur Admins → § 7.5), **Abmelden** |
+| **⚙ Admin-Bereich** | Öffnet das Admin-Panel — nur für Admin-Nutzer sichtbar (→ § 7.5) |
 
 **Unten:**
 
diff --git a/app/web/index.html b/app/web/index.html
index 7750b84..f0c5a89 100644
--- a/app/web/index.html
+++ b/app/web/index.html
@@ -233,6 +233,12 @@
       class="h-9 w-9 grid place-items-center rounded-lg border border-slate-300 dark:border-slate-600 hover:bg-slate-100 dark:hover:bg-slate-700 text-xl leading-none transition-colors">⋮</button>
   </header>
 
+  <!-- Identitätsleiste: angemeldeter Nutzer (links) + Abmelden (rechts) -->
+  <div class="shrink-0 flex items-center justify-between px-4 py-1 text-xs text-slate-500 dark:text-slate-400 border-b border-slate-200 dark:border-slate-700">
+    <span id="identity">lade …</span>
+    <a id="logout" class="hidden hover:underline" href="#">Abmelden</a>
+  </div>
+
   <!-- Verstecktes Quell-Select: speist die Ton-Presets (bewahrt die bestehende Logik) -->
   <select id="tts" class="hidden">
     <option value="device">Im Gerät</option>
@@ -294,16 +300,11 @@
         <span id="theme-icon" class="w-5 text-center shrink-0">🌙</span>
         <span>Tag-/Nachtmodus</span>
       </button>
-    </div>
-
-    <!-- Konto -->
-    <div class="border-t border-slate-100 dark:border-slate-700 pt-2">
-      <p id="identity" class="text-xs text-slate-500 dark:text-slate-400 px-1 mb-1.5">lade …</p>
-      <div class="flex items-center gap-2 px-1">
-        <button id="admin-toggle" type="button"
-          class="hidden text-xs rounded-lg border border-slate-300 dark:border-slate-600 px-2.5 py-1.5 hover:bg-slate-100 dark:hover:bg-slate-700 transition-colors">⚙ Admin</button>
-        <a id="logout" href="#" class="hidden text-xs text-blue-600 dark:text-blue-400 hover:underline ml-auto">Abmelden</a>
-      </div>
+      <button id="admin-toggle" type="button"
+        class="hidden w-full flex items-center gap-3 rounded-lg px-3 py-2 text-sm hover:bg-slate-50 dark:hover:bg-slate-700/50 text-left transition-colors">
+        <span class="w-5 text-center shrink-0">⚙</span>
+        <span>Admin-Bereich</span>
+      </button>
     </div>
   </div>
 
@@ -325,6 +326,6 @@
     <div id="status" class="w-full max-w-3xl mx-auto mt-1 min-h-[1rem] text-xs text-slate-500 dark:text-slate-400"></div>
   </footer>
 
-  <script src="/app.js?v=40"></script>
+  <script src="/app.js?v=41"></script>
 </body>
 </html>

From b8b2bcddd3c838345d93d0cda057df280f89635a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Dieter=20Schl=C3=BCter?= <dieter.schlueter@linix.de>
Date: Sun, 21 Jun 2026 01:51:37 +0200
Subject: [PATCH 03/10] =?UTF-8?q?ui(tts):=20"Im=20Ger=C3=A4t"=20ausgrauen?=
 =?UTF-8?q?=20statt=20ausblenden,=20wenn=20keine=20Sprachstimmen?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- refreshDevicePreset(): graut das "Im Gerät"-Preset aus (disabled + opacity + Hinweis
  "keine Sprachstimmen im Browser"), wenn getVoices() leer ist; entgraut automatisch
  bei voiceschanged (+ 1,5s-Backstop).
- Kein Aufzwingen: device wird nur (wieder) gewählt, wenn es die gespeicherte Präferenz
  ist; ohne Stimmen wird die aktive Wahl auf piper gesetzt (Präferenz bleibt erhalten).
- Server-Fallback bleibt für fehlende Einzelsprachen.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 app/web/app.js     | 35 +++++++++++++++++++++++++++--------
 app/web/index.html |  4 ++--
 2 files changed, 29 insertions(+), 10 deletions(-)

diff --git a/app/web/app.js b/app/web/app.js
index d3d6ae5..a7ce676 100644
--- a/app/web/app.js
+++ b/app/web/app.js
@@ -60,7 +60,9 @@ let _voices = [];
 function loadVoices() { _voices = TTS_SUPPORTED ? speechSynthesis.getVoices() : []; }
 if (TTS_SUPPORTED) {
   loadVoices();
-  speechSynthesis.addEventListener("voiceschanged", loadVoices);
+  // Stimmen laden asynchron -> bei Eintreffen Preset-Zustand neu bewerten (entgrauen).
+  speechSynthesis.addEventListener("voiceschanged", () => { loadVoices(); refreshDevicePreset(); });
+  setTimeout(() => refreshDevicePreset(), 1500);  // Backstop, falls kein voiceschanged kommt
 }
 
 function pickVoice(bcp) {
@@ -139,13 +141,6 @@ const PLAYBACK_KEY = "va-playback";  // geräte-lokal: "device" wenn Geräte-TTS
 // Vorrang: 1. lokale Wahl (localStorage), 2. Server-Pref (tts_provider), 3. Mobil-Default = Gerät.
 function applyPlaybackChoice(serverProvider) {
   if (!ttsSel) return;
-  // "Im Gerät"-Preset ausblenden, wenn der Browser keine Web Speech API hat.
-  if (!TTS_SUPPORTED) {
-    const opt = ttsSel.querySelector('option[value="device"]');
-    if (opt) opt.remove();
-    const btn = document.querySelector('.tts-preset[data-tts="device"]');
-    if (btn) btn.classList.add("hidden");
-  }
   const local = localStorage.getItem(PLAYBACK_KEY);
   if (local === "device" && TTS_SUPPORTED) {
     ttsSel.value = "device";
@@ -157,6 +152,30 @@ function applyPlaybackChoice(serverProvider) {
   } else if (!ttsSel.value) {
     ttsSel.value = "piper";               // Fallback: Server
   }
+  refreshDevicePreset();                  // graut „Im Gerät" aus, wenn keine Stimmen + syncPresetUI
+}
+
+// Graut „Im Gerät" aus, wenn keine Sprachstimmen vorhanden sind; entgraut automatisch,
+// sobald welche da sind. Kein Aufzwingen: device wird nur (wieder) gewählt, wenn es die
+// gespeicherte Präferenz ist. Ohne Stimmen NICHT auf der deaktivierten Option sitzen bleiben.
+function refreshDevicePreset() {
+  const btn = document.querySelector('.tts-preset[data-tts="device"]');
+  if (!btn || !ttsSel) return;
+  const ready = deviceVoicesReady();
+  btn.disabled = !ready;
+  btn.classList.toggle("opacity-50", !ready);
+  btn.classList.toggle("cursor-not-allowed", !ready);
+  const sub = btn.querySelector(".tts-sub");
+  if (sub) sub.textContent = ready
+    ? "Das Gerät liest vor · spart Daten"
+    : "keine Sprachstimmen im Browser";
+  if (ready) {
+    if (localStorage.getItem(PLAYBACK_KEY) === "device" && ttsSel.value !== "device") {
+      ttsSel.value = "device";            // gespeicherte Präferenz wieder aktivieren
+    }
+  } else if (ttsSel.value === "device") {
+    ttsSel.value = "piper";               // Server statt deaktivierter Option (Präferenz bleibt)
+  }
   syncPresetUI();
 }
 
diff --git a/app/web/index.html b/app/web/index.html
index f0c5a89..fafb5b0 100644
--- a/app/web/index.html
+++ b/app/web/index.html
@@ -261,7 +261,7 @@
           class="tts-preset w-full flex items-center gap-3 rounded-lg border border-slate-200 dark:border-slate-700 px-3 py-2 text-left hover:bg-slate-50 dark:hover:bg-slate-700/50 transition-colors">
           <span class="text-lg shrink-0">📱</span>
           <span class="flex-1 min-w-0"><span class="block text-sm font-medium">Im Gerät</span>
-            <span class="block text-xs text-slate-500 dark:text-slate-400">Das Gerät liest vor · spart Daten</span></span>
+            <span class="tts-sub block text-xs text-slate-500 dark:text-slate-400">Das Gerät liest vor · spart Daten</span></span>
           <span class="check text-blue-600 dark:text-blue-400 font-bold opacity-0">✓</span>
         </button>
         <button type="button" data-tts="piper"
@@ -326,6 +326,6 @@
     <div id="status" class="w-full max-w-3xl mx-auto mt-1 min-h-[1rem] text-xs text-slate-500 dark:text-slate-400"></div>
   </footer>
 
-  <script src="/app.js?v=41"></script>
+  <script src="/app.js?v=42"></script>
 </body>
 </html>

From 303372f0b96e6965cd2af6d20669bd4019c82c2b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Dieter=20Schl=C3=BCter?= <dieter.schlueter@linix.de>
Date: Sun, 21 Jun 2026 15:27:34 +0200
Subject: [PATCH 04/10] =?UTF-8?q?docs:=20Admin-API=20(=C2=A77.1),=20Admin-?=
 =?UTF-8?q?Panel/Ger=C3=A4te-TTS/Presets=20+=20Verzeichnisstruktur=20nachz?=
 =?UTF-8?q?iehen?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 Docs/voice-assistant-architecture.md | 48 +++++++++++++++++++++++++---
 README.md                            |  5 +--
 2 files changed, 46 insertions(+), 7 deletions(-)

diff --git a/Docs/voice-assistant-architecture.md b/Docs/voice-assistant-architecture.md
index 66b5315..580a247 100644
--- a/Docs/voice-assistant-architecture.md
+++ b/Docs/voice-assistant-architecture.md
@@ -170,7 +170,6 @@ ein No-op; `LoopbackOutput` sammelt die Chunks (testbar ohne Hardware).
 | `GET  /api/devices`                  | verfügbare Audio-Endpunkte + Capabilities |
 | `POST /api/sessions/{id}/route`      | bevorzugte Geräte/Provider/Sprache je Session |
 | `GET  /api/config`                   | aktives Profil + aufgelöste Route (ohne Secrets) |
-| `POST /api/admin/users`              | Nutzer anlegen (Admin-Key) → Token einmalig |
 | `GET  /api/me` · `PUT /api/me/prefs` | aktueller Nutzer + dauerhafte Präferenzen |
 | `GET/POST/DELETE /api/me/memories`   | Langzeit-Erinnerungen des Nutzers |
 | `GET  /api/metrics`                  | Metriken (JSON / Prometheus) |
@@ -181,6 +180,32 @@ Endpunkt-/Provider-Auswahl ist über **Request-Body** (pro Aufruf), **Session**
 (`?session_id=…`) und **Defaults/Profil** steuerbar. Verwendete Route erscheint als
 `X-*`-Header bzw. im `?debug`-JSON.
 
+### 7.1 Admin-API (alle hinter `require_admin`, Audit-geloggt)
+
+Trägt das Admin-Web-Panel (5 Bereiche + Übersicht-Dashboard). Schreibende Aktionen
+werden ins Audit-Log geschrieben.
+
+| Methode & Pfad                                   | Zweck |
+|--------------------------------------------------|-------|
+| `POST   /api/admin/users`                        | Nutzer anlegen → Token einmalig |
+| `GET    /api/admin/users`                        | Nutzerliste |
+| `PUT/DELETE /api/admin/users/{id}`               | Nutzer ändern/löschen |
+| `POST   /api/admin/users/{id}/token`             | Token neu ausstellen |
+| `GET/POST/DELETE /api/admin/users/{id}/memories` | Erinnerungen je Nutzer |
+| `GET    /api/admin/users/{id}/sessions`·`/usage` | Sessions / Kontingent-Nutzung |
+| `GET    /api/admin/sessions/{id}/messages`       | Gesprächsverlauf einsehen |
+| `GET/PUT/DELETE /api/admin/config[/{key}]`       | Live-Config lesen/setzen (z. B. `top_p`) |
+| `GET    /api/admin/llm/status`                   | LLM-/GPU-Status (read-only) |
+| `POST   /api/admin/llm/backend`                  | Backend wechseln (Ollama ↔ llama.cpp) |
+| `POST   /api/admin/gateway/restart`              | Gateway aus dem Panel neu starten |
+| `GET/POST/DELETE /api/admin/pronunciation/{lang}`| Aussprache-Lexika pflegen |
+| `GET    /api/admin/usage`·`/emergency-events`    | Gesamt-Nutzung / Notfall-Ereignisse |
+| `GET    /api/admin/db-export`                    | SQLite-Export |
+| `WS     /api/admin/log`                          | Live-Log-Stream |
+
+Config-Änderungen sind als **Live** (sofort wirksam) oder **Restart** (Neustart nötig)
+gekennzeichnet; der Backend-Wechsel und Live-Parameter wie `top_p` laufen ohne Neustart.
+
 ## 8. Stand der Implementierung
 
 **Umgesetzt:** FastAPI-Gateway, alle o. g. REST-Endpunkte; OpenRouter-Adapter für
@@ -195,7 +220,15 @@ Store, fließt ins LLM)**; **Langzeit-Erinnerungen pro Nutzer (als LLM-Kontext)*
 über WebSocket (`/ws/voice`: Audio rein → STT → Antwort-Pipeline) mit VAD-Aeusserungs-
 erkennung und Barge-in (`interrupt`)**; **Resilienz (Fallback-Ketten je Modul,
 In-Memory-Metriken `/api/metrics`), Tageskontingent pro Nutzer und heuristische
-Notfall-Eskalation**; automatisierte Tests.
+Notfall-Eskalation**; **Admin-Web-Panel (5 Bereiche + Übersicht-Dashboard) über die
+Admin-API (§7.1) inkl. Backend-Wechsel, Gateway-Neustart, Live-Config-Parameter,
+LLM-/GPU-Status, Aussprache-Lexika, Live-Log und Audit-Logging schreibender
+Aktionen**; automatisierte Tests.
+
+**Web-/Mobil-Frontend:** schlankes Web-Interface unter `/` (Tailwind, kein Build),
+mit **Geräte-TTS** (Browser-SpeechSynthesis auf Mobilgeräten; fällt auf Server-Audio
+zurück, wenn keine lokalen Stimmen vorhanden) und **Ton-Presets** (Schnell → piper,
+Hohe Qualität → chatterbox, Cloud → openrouter). Auth-Gate (Bearer-Token).
 
 **Echtes lokales STT & TTS:** `faster-whisper` (optionale Dependency `.[local]`,
 CTranslate2) transkribiert real; `piper` (in-process via piper-Python-API, Stimmmodell
@@ -235,16 +268,21 @@ voice-assistant-scaffold/
 │   ├── dependencies.py    # Registries, ResolvedRoute, resolve_route, Store-/Router-Singleton
 │   ├── store.py           # Persistenz: Store-Interface + SQLiteStore (Nutzer/Sessions/Verlauf)
 │   ├── auth.py            # Bearer-Token-Auth (require_user) + Admin-Schutz
+│   ├── audit.py           # Audit-Log schreibender Admin-Aktionen
+│   ├── admin_llm.py       # LLM-/GPU-Status + Backend-Wechsel (Ollama ↔ llama.cpp)
+│   ├── runtime_config.py  # Live-Config (zur Laufzeit setzbare Parameter)
 │   ├── metrics.py         # In-Memory-Metriken (Counter/Timer, JSON + Prometheus)
 │   ├── quota.py           # Tageskontingent pro Nutzer (Kostenkontrolle)
-│   ├── safety/            # emergency.py: heuristische Notfall-Erkennung/-Eskalation
+│   ├── safety/            # emergency.py (Heuristik) + llm_classifier.py (Notfall-Klassifikation)
 │   ├── errors.py          # RoutingError -> HTTP 422
 │   ├── schemas.py         # Pydantic-Modelle
 │   ├── api/               # health, chat, speak, transcribe, devices, sessions, config, admin, me, ws
-│   ├── core/              # orchestrator
+│   ├── core/              # orchestrator, memory_extractor (Auto-Erinnerungen), warmup (Modell-Vorladen)
 │   ├── audio/             # router, transport_router, vad, endpoints/input|output/*
 │   ├── pipeline/          # input_cleaner, spoken_response_adapter, tts_normalizer, sentence_chunker, german_numbers
-│   └── providers/         # stt/ llm/ tts/ (openrouter + lokale Stubs) + fallback.py
+│   ├── providers/         # stt/ llm/ tts/ (openrouter + lokale + chatterbox) + fallback.py
+│   ├── utils/             # Hilfsfunktionen
+│   └── web/               # Web-/Admin-Frontend (Tailwind, kein Build)
 ├── config/                # voice-assistant.example.toml (+ lokale .toml, gitignored)
 ├── data/                  # SQLite-DB (gitignored)
 ├── deploy/                # systemd unit + env-Beispiel
diff --git a/README.md b/README.md
index fc1fc9f..d28ed79 100644
--- a/README.md
+++ b/README.md
@@ -17,7 +17,8 @@ Jede Achse — Hardware, Betrieb, Software — ist frei konfigurierbar, ohne Cod
 - **Gesprächsgedächtnis** pro Session (Verlauf) + Langzeit-Erinnerungen pro Nutzer (manuell + automatisch)
 - **WebSocket-Streaming:** Token-Streaming (LLM), Audio-Streaming (satzweises TTS), Sprach-Eingang, VAD, Barge-in
 - **Notfall-Eskalation:** zweistufig (Stichwörter + LLM-Klassifikation im Hintergrund)
-- **Web-Interface** unter `/` (Tailwind, kein Build, mobiltauglich)
+- **Web-Interface** unter `/` (Tailwind, kein Build, mobiltauglich) mit Geräte-TTS (Browser-Sprachausgabe) und Ton-Presets (Schnell/Hohe Qualität/Cloud)
+- **Admin-Panel** (5 Bereiche + Übersicht-Dashboard): Nutzer/Sessions, LLM-/GPU-Status, Backend-Wechsel + Gateway-Neustart, Live-Config, Aussprache-Lexika, Live-Log, Audit-Logging
 - **Keine Secrets im Code** — API-Keys nur über die Umgebung
 
 ---
@@ -193,7 +194,7 @@ make llm-status # Container- und HTTP-Status prüfen
 ## Projektstruktur (Kurzform)
 
 ```text
-app/         Gateway: config, api/, core/, audio/, pipeline/, providers/
+app/         Gateway: config, api/, core/, audio/, pipeline/, providers/, web/ (Frontend + Admin)
 config/      voice-assistant.example.toml  (lokale .toml ist gitignored)
 deploy/      systemd-Unit, nginx-Vorlage, env-Beispiel
 scripts/     voice_loop.py, chat_client.py, smoke_e2e.py, add_pronunciation.py

From 789e3510fc6b14a470a7f3a041989d009e67bc26 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Dieter=20Schl=C3=BCter?= <dieter.schlueter@linix.de>
Date: Sun, 21 Jun 2026 15:56:33 +0200
Subject: [PATCH 05/10] chore: Pfade auf ~/my_voice_assistant_v3 umstellen
 (Repo-Umzug v2 -> v3)

systemd-Unit-Vorlage + Doku-Pfade auf den neuen flachen Pfad angepasst.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 BEDIENUNGSANLEITUNG.md               | 2 +-
 Docs/voice-assistant-architecture.md | 2 +-
 deploy/voice-assistant.user.service  | 4 ++--
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/BEDIENUNGSANLEITUNG.md b/BEDIENUNGSANLEITUNG.md
index f03cbb0..2a0aa62 100644
--- a/BEDIENUNGSANLEITUNG.md
+++ b/BEDIENUNGSANLEITUNG.md
@@ -135,7 +135,7 @@ pip install -e .[local]   # installiert faster-whisper + piper-tts
 ### 2.2 Installation
 
 ```bash
-cd voice-assistant-scaffold
+cd my_voice_assistant_v3
 python3 -m venv .venv
 source .venv/bin/activate
 pip install -U pip
diff --git a/Docs/voice-assistant-architecture.md b/Docs/voice-assistant-architecture.md
index 580a247..15b3e49 100644
--- a/Docs/voice-assistant-architecture.md
+++ b/Docs/voice-assistant-architecture.md
@@ -261,7 +261,7 @@ Verschlüsselung at-rest/in-transit, Löschkonzept, Einwilligung — „privacy
 ## 10. Verzeichnisstruktur (Ist)
 
 ```text
-voice-assistant-scaffold/
+my_voice_assistant_v3/
 ├── app/
 │   ├── main.py            # FastAPI-App + Router-Registrierung
 │   ├── config.py          # Settings, TOML-Profile, Präzedenz
diff --git a/deploy/voice-assistant.user.service b/deploy/voice-assistant.user.service
index 2fab452..5d54725 100644
--- a/deploy/voice-assistant.user.service
+++ b/deploy/voice-assistant.user.service
@@ -18,8 +18,8 @@ After=network.target
 
 [Service]
 Type=simple
-WorkingDirectory=/home/dschlueter/my_voice_assistant_v2/voice-assistant-scaffold
-ExecStart=/home/dschlueter/my_voice_assistant_v2/voice-assistant-scaffold/.venv/bin/uvicorn app.main:app --host 0.0.0.0 --port 8003
+WorkingDirectory=/home/dschlueter/my_voice_assistant_v3
+ExecStart=/home/dschlueter/my_voice_assistant_v3/.venv/bin/uvicorn app.main:app --host 0.0.0.0 --port 8003
 Restart=always
 RestartSec=2
 

From 8bee8dce581297600ae699617c135c9fae02c9b1 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Dieter=20Schl=C3=BCter?= <dieter.schlueter@linix.de>
Date: Sun, 21 Jun 2026 16:35:55 +0200
Subject: [PATCH 06/10] docs(handbuch): curl-Beispiele vereinheitlichen (| jq,
 -s, $URL, X-Admin-Key)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- JSON-liefernde curl-Befehle ohne `| jq` ergaenzt (lesbare Ausgabe)
- alle `| jq`-Befehle nutzen jetzt einheitlich `curl -s`
- §6.5-Block an den Hausstil angeglichen:
  - `http://localhost:8080` -> `$URL` (zog frueher den falschen Port; real 8003)
  - irrefuehrendes `Authorization: Bearer <admin-token>` -> `X-Admin-Key: $ADMIN_API_KEY`
    (require_admin akzeptiert kein generisches Admin-Token; User-Token => 401)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 BEDIENUNGSANLEITUNG.md | 28 ++++++++++++++--------------
 1 file changed, 14 insertions(+), 14 deletions(-)

diff --git a/BEDIENUNGSANLEITUNG.md b/BEDIENUNGSANLEITUNG.md
index 2a0aa62..c396874 100644
--- a/BEDIENUNGSANLEITUNG.md
+++ b/BEDIENUNGSANLEITUNG.md
@@ -647,7 +647,7 @@ VA_PROFILE=hybrid make run          # oder: systemctl --user restart voice-assis
 
 **Prüfen** (egal welche Richtung):
 ```bash
-curl http://127.0.0.1:11434/v1/models   # Ollama   (bzw. :8001 für llama.cpp)
+curl -s http://127.0.0.1:11434/v1/models | jq   # Ollama   (bzw. :8001 für llama.cpp)
 # danach im Admin → Status den LLM-Provider/das Modell kontrollieren oder kurz testen
 ```
 
@@ -1324,18 +1324,18 @@ Der Cache wird automatisch geleert.
 **Weg 2 — REST-API** (alle Sprachen, sofort wirksam):
 ```bash
 # Französischen Eintrag hinzufügen (kein Neustart nötig):
-curl -X POST http://localhost:8080/api/admin/pronunciation/fr \
-  -H "Authorization: Bearer <admin-token>" \
+curl -s -X POST $URL/api/admin/pronunciation/fr \
+  -H "X-Admin-Key: $ADMIN_API_KEY" \
   -H "Content-Type: application/json" \
-  -d '{"section":"terms","key":"Schlüter","value":"Chluteur"}'
+  -d '{"section":"terms","key":"Schlüter","value":"Chluteur"}' | jq
 
 # Eintrag löschen:
-curl -X DELETE http://localhost:8080/api/admin/pronunciation/fr/terms/Schlüter \
-  -H "Authorization: Bearer <admin-token>"
+curl -s -X DELETE $URL/api/admin/pronunciation/fr/terms/Schlüter \
+  -H "X-Admin-Key: $ADMIN_API_KEY" | jq
 
 # Alle Einträge einer Sprache anzeigen:
-curl http://localhost:8080/api/admin/pronunciation/ru \
-  -H "Authorization: Bearer <admin-token>"
+curl -s $URL/api/admin/pronunciation/ru \
+  -H "X-Admin-Key: $ADMIN_API_KEY" | jq
 ```
 
 **Weg 3 — YAML-Datei direkt editieren** (alle Sprachen):
@@ -1356,7 +1356,7 @@ Spricht den Testsatz mit der aktuell eingestellten Stimme und Sprache.
 
 **Oder via curl:**
 ```bash
-curl -s -X POST http://localhost:8080/api/speak \
+curl -s -X POST $URL/api/speak \
   -H "Authorization: Bearer $TOKEN" \
   -H "Content-Type: application/json" \
   -d '{"text":"Hallo, ich bin Dieter Schlüter.","tts_provider":"piper","language":"fr"}' \
@@ -1411,12 +1411,12 @@ DEFAULT_LANGUAGE_MODE=fix  # global: fix | flex (Admin: Tab „⚙ Einstellungen
 Pro Nutzer (dauerhaft):
 ```bash
 # Feste Sprache (Fix):
-curl -X PUT $URL/api/me/prefs -H "Authorization: Bearer $TOKEN" \
-  -H 'Content-Type: application/json' -d '{"language":"en","language_mode":"fix"}'
+curl -s -X PUT $URL/api/me/prefs -H "Authorization: Bearer $TOKEN" \
+  -H 'Content-Type: application/json' -d '{"language":"en","language_mode":"fix"}' | jq
 
 # Flex (Sprache folgt automatisch):
-curl -X PUT $URL/api/me/prefs -H "Authorization: Bearer $TOKEN" \
-  -H 'Content-Type: application/json' -d '{"language_mode":"flex"}'
+curl -s -X PUT $URL/api/me/prefs -H "Authorization: Bearer $TOKEN" \
+  -H 'Content-Type: application/json' -d '{"language_mode":"flex"}' | jq
 ```
 
 Pro Aufruf: `{"text":"…","language":"en","language_mode":"fix"}` im Body.
@@ -1897,7 +1897,7 @@ curl -s -X POST $URL/api/me/memories \
 curl -s $URL/api/me/memories -H "Authorization: Bearer $TOKEN" | jq
 
 # Erinnerung löschen:
-curl -s -X DELETE $URL/api/me/memories/<id> -H "Authorization: Bearer $TOKEN"
+curl -s -X DELETE $URL/api/me/memories/<id> -H "Authorization: Bearer $TOKEN" | jq
 ```
 
 In der Datenbank direkt ansehen/löschen (nötig z. B. wenn eine falsch extrahierte

From e99508a727194b383ccb93996e4b6a677252506c Mon Sep 17 00:00:00 2001
From: dschlueter <dschlueter@kitux.de>
Date: Sun, 21 Jun 2026 17:22:36 +0200
Subject: [PATCH 09/10] docs: CLAUDE.md fuer Claude Code ergaenzt (Befehle +
 Architektur-Ueberblick)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 CLAUDE.md | 69 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 69 insertions(+)
 create mode 100644 CLAUDE.md

diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 0000000..4eb982d
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,69 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+> Sprache: Code-Kommentare, Doku und Commits sind auf Deutsch. Antworte und kommentiere auf Deutsch.
+
+## Befehle
+
+```bash
+make install          # venv anlegen + pip install -e .[test]
+make run              # Gateway (uvicorn --reload, Port aus .env, Standard 8080)
+make test             # gesamte Pytest-Suite (offline, ohne Netz/Kosten)
+make smoke            # echter End-to-End-Test gegen OpenRouter (kostet wenig)
+make stop / restart   # Gateway + llama.cpp stoppen / neu starten
+
+# Einzelner Test / einzelner Fall:
+. .venv/bin/activate && pytest tests/test_routing.py
+. .venv/bin/activate && pytest tests/test_routing.py::test_name -q
+```
+
+Profil beim Start wählen (überschreibt `.env`): `VA_PROFILE=cloud|hybrid|local-dev make run`.
+Lokales LLM (llama.cpp-Docker) für `hybrid`/`local-dev`: `make llm-up` → `make llm-status` → `make llm-down`.
+Backend wechseln (llama.cpp ↔ Ollama, teilen sich die GPU, nie gleichzeitig): `make llm-llamacpp` / `make llm-ollama`.
+
+Vor `make run`: wenn `.env` fehlt, wird sie aus `.env.example` erzeugt. Für Cloud/Hybrid `OPENROUTER_API_KEY` setzen.
+
+## Architektur (großes Bild)
+
+Modulares FastAPI-Gateway für einen seniorengerechten Sprachassistenten. Leitprinzip: **jede Achse (STT/LLM/TTS, Audio-I/O, lokal vs. remote) austauschbar, ohne Kern-Code zu ändern.** Ausführliche Referenz: `Docs/voice-assistant-architecture.md` und `BEDIENUNGSANLEITUNG.md`.
+
+### Zwei zentrale Mechanismen, die man verstehen muss
+
+1. **Geschichtete Konfiguration mit Profilen** (`app/config.py`). Präzedenz, höhere Ebene gewinnt:
+   `Defaults < config/voice-assistant.toml ([defaults]+[profiles.<aktiv>]) < ENV/.env < Nutzer-Prefs < Session-Route < Request`.
+   `VA_PROFILE` aktiviert ein Profil; `TomlProfileSource` hängt die TOML-Werte als pydantic-settings-Quelle **unter** ENV ein. **Secrets nie in die TOML** — nur in die Umgebung. Profile: `cloud` (alles OpenRouter), `hybrid` (LLM lokal), `local-dev` (alles lokal).
+
+2. **Route-Auflösung + Registries** (`app/dependencies.py`). `resolve_route(user, session_id, overrides)` liefert pro Aufruf eine `ResolvedRoute` (Endpunkte, STT/LLM/TTS-Provider, Sprache) nach obiger Präzedenz. Provider kommen aus `STT_REGISTRY`/`LLM_REGISTRY`/`TTS_REGISTRY` (`name -> factory(settings)`). **Neuen Provider = ein Registry-Eintrag + ABC-Implementierung** in `app/providers/{stt,llm,tts}/`. Unbekannter Name → `UnknownComponentError` → HTTP 422.
+
+### Speech-Pipeline (`app/pipeline/`, orchestriert in `app/core/orchestrator.py`)
+
+Kernidee: **semantische** Antwort ≠ **gesprochene** Antwort. Stufen:
+`raw_transcript → cleaned_transcript (input_cleaner) → semantic_response (LLM) → spoken_response (spoken_response_adapter) → tts_ready_text (tts_normalizer) → TTS`.
+
+- `spoken_response_adapter`: macht LLM-Text sprechbar (Markdown raus, Listen → Ordinalwörter).
+- `tts_normalizer`: füllt nur Phonemizer-Lücken (Ordinalia, Einheiten, Abkürzungen, YAML-Lexikon `config/pronunciation.<lang>.yaml`). Stufe provider-abhängig (`TTS_NORMALIZE_LEVEL`): piper→`full`, Cloud-TTS→`light`. Deutsche Ordinalzahlen in `german_numbers.py`.
+- `sentence_chunker`: inkrementelle Satzsegmentierung fürs satzweise Streaming-TTS; trennt bewusst **nicht** bei „1. Mai", „z. B." usw.
+
+Der Orchestrator schreibt synthetisiertes Audio **zusätzlich** in den Output-Endpunkt (`open→write_chunk→flush→close`) **und** gibt es als HTTP-Stream zurück. Bei `/api/chat` mit `session_id` lädt die API-Schicht den Verlauf aus dem Store, gibt ihn als `history` ans LLM und speichert beide Turns; ohne `session_id` zustandslos.
+
+### Querschnitt
+
+- **Persistenz** (`app/store.py`): `Store`-Interface + `SQLiteStore` (Nutzer, Sessions, Verlauf, Erinnerungen) — DB in `data/` (gitignored).
+- **Auth** (`app/auth.py`): Bearer-Token (`require_user`) + Admin-Schutz (`require_admin`). `main.py` hat zwei Middlewares: Metriken + Web-UI-Gate (statische Seiten hinter Auth/SSO; `/api`, `/ws`, `/health` ausgenommen). `AUTH_ENABLED=false` → anonymer Nutzer (LAN-Dev).
+- **Resilienz** (`app/providers/fallback.py`): Fallback-Ketten je Modul (`*_FALLBACK`). Metriken in `app/metrics.py` (`/api/metrics`, JSON + Prometheus). Tageskontingent `app/quota.py` (429).
+- **Sicherheit** (`app/safety/`): zweistufige Notfall-Eskalation — schnelle Stichwort-Heuristik im Hot-Path (`emergency.py`) + LLM-Klassifikation als nicht-blockierender Hintergrund-Task (`llm_classifier.py`).
+- **Auto-Erinnerungen** (`app/core/memory_extractor.py`): nach N Turns destilliert ein LLM dauerhafte Fakten als Hintergrund-Task.
+- **Echtzeit** (`app/api/ws.py`): `/ws/chat` (Token-/Audio-Streaming) und `/ws/voice` (Audio→STT→Pipeline, VAD, Barge-in via `interrupt`).
+- **Admin** (`app/api/admin.py`, `app/admin_llm.py`, `app/audit.py`, `app/runtime_config.py`): Live-Config, Backend-Wechsel, Gateway-Neustart, Lexika, Live-Log — schreibende Aktionen werden ins Audit-Log geschrieben.
+- **Frontend** (`app/web/`): Tailwind ohne Build-Schritt, mobiltauglich. Geräte-TTS (Browser-SpeechSynthesis) mit Fallback auf Server-Audio.
+
+### Bewusste Platzhalter (nicht „kaputt")
+
+Audio-Endpunkte (`app/audio/endpoints/`) liefern leere Chunks — nur Auswahl/Lifecycle sind verdrahtet, kein echtes Hardware-I/O. `app/audio/transport_router.py` (Ebene 4 lokal/remote) existiert, ist aber noch nicht aktiv; lokal/remote trägt vorerst der Provider-Name.
+
+## Provider & lokale Module
+
+- Cloud: OpenRouter (STT/LLM/TTS). Lokales LLM: OpenAI-kompatibel (`local_openai_compatible.py`) gegen llama.cpp oder Ollama.
+- Lokale KI ist **optional**: `pip install -e .[local]` (faster-whisper, piper). Beim Serverstart werden lokale Modelle vorgeladen (`app/core/warmup.py`).
+- `chatterbox`-TTS (eigener GPU-HTTP-Dienst, Voice-Cloning) ist als wählbarer Provider angebunden (job-basiert `/speak`→`/status`→`/audio`).

From 91b9ef3374eb3908fa96c8662de7a8bdf4e5f37f Mon Sep 17 00:00:00 2001
From: dschlueter <dschlueter@kitux.de>
Date: Wed, 24 Jun 2026 22:01:38 +0200
Subject: [PATCH 10/10] =?UTF-8?q?Jamulix:=20Optimierungen=20=C3=BCbernehme?=
 =?UTF-8?q?n,=20lokale=20Voice-Modelle=20ignorieren?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .gitignore                      |   4 +
 app/api/chat.py                 |   3 +-
 app/api/config.py               |  58 ++++++++++-
 app/api/speak.py                |   3 +-
 app/api/ws.py                   |   9 +-
 app/config.py                   |   5 +
 app/dependencies.py             |  65 +++++++++++--
 app/providers/llm/openrouter.py | 167 +++++++++++++++++++++-----------
 app/providers/tts/cartesia.py   | 142 +++++++++++++++++++++++++++
 app/providers/tts/piper.py      |  30 ++++--
 app/web/app.js                  | 117 +++++++++++++++++++++-
 app/web/index.html              |  34 ++++++-
 12 files changed, 549 insertions(+), 88 deletions(-)
 create mode 100644 app/providers/tts/cartesia.py

diff --git a/.gitignore b/.gitignore
index e00fe4c..479d860 100644
--- a/.gitignore
+++ b/.gitignore
@@ -32,3 +32,7 @@ venv/
 3
 TODO.md
 
+
+# Local voice models
+config/voices/*.onnx
+config/voices/*.onnx.json
diff --git a/app/api/chat.py b/app/api/chat.py
index f56a0db..b0a3fa6 100644
--- a/app/api/chat.py
+++ b/app/api/chat.py
@@ -13,6 +13,7 @@ from app.dependencies import (
     resolve_output_endpoint,
     get_store,
     piper_voice_for_language,
+    voice_for_route,
 )
 from app.core.memory_extractor import maybe_schedule_extraction
 from app.quota import enforce_quota, record_usage, QuotaExceededError
@@ -96,7 +97,7 @@ async def chat(
             raise HTTPException(status_code=429, detail=str(exc))
 
     # Explizit angeforderte Stimme gewinnt; sonst folgt sie der Routensprache.
-    voice = payload.voice or piper_voice_for_language(route.tts_provider, route.language)
+    voice = payload.voice or voice_for_route(route.tts_provider, route.language, route.voice_gender)
 
     try:
         trace, audio = await orchestrator.chat_text(
diff --git a/app/api/config.py b/app/api/config.py
index 6b1f1e5..c939955 100644
--- a/app/api/config.py
+++ b/app/api/config.py
@@ -1,3 +1,4 @@
+import httpx
 from fastapi import APIRouter
 
 from app.config import settings, active_profile
@@ -12,14 +13,60 @@ from app.dependencies import (
 router = APIRouter()
 
 
+async def _chatterbox_reachable() -> bool:
+    try:
+        async with httpx.AsyncClient(timeout=1.5) as client:
+            r = await client.get(f"{settings.chatterbox_base_url.rstrip('/')}/status")
+            return r.status_code < 500
+    except Exception:
+        return False
+
+
+async def _cartesia_available() -> bool:
+    key = settings.cartesia_api_key.strip()
+    voice_id = settings.cartesia_voice_id.strip()
+    if not key or not voice_id:
+        return False
+    try:
+        async with httpx.AsyncClient(timeout=5.0) as client:
+            r = await client.get(
+                "https://api.cartesia.ai/voices",
+                headers={"X-API-Key": key, "Cartesia-Version": "2024-06-10"},
+            )
+            return r.status_code < 400
+    except Exception:
+        return False
+
+
+async def _openrouter_tts_available() -> bool:
+    """Prüft ob das konfigurierte TTS-Modell auf OpenRouter via /audio/speech erreichbar ist."""
+    key = settings.openrouter_api_key.strip()
+    model = settings.openrouter_tts_model.strip()
+    if not key or not model:
+        return False
+    try:
+        async with httpx.AsyncClient(timeout=5.0) as client:
+            r = await client.post(
+                "https://openrouter.ai/api/v1/audio/speech",
+                headers={"Authorization": f"Bearer {key}", "Content-Type": "application/json"},
+                json={"model": model, "input": "Test", "voice": settings.openrouter_tts_voice or "alloy",
+                      "response_format": "mp3"},
+            )
+            # 200 = Audio OK, 4xx mit "does not exist" = Modell fehlt
+            if r.status_code == 200:
+                return True
+            body = r.text
+            return "does not exist" not in body and r.status_code < 500
+    except Exception:
+        return False
+
+
 @router.get("/config")
 async def get_config():
-    """Zeigt aktives Profil, aufgeloeste Default-Route und verfuegbare Bausteine.
-
-    Bewusst OHNE Secrets - API-Keys werden nur als 'gesetzt/nicht gesetzt' gemeldet.
-    """
+    """Zeigt aktives Profil, aufgeloeste Default-Route und verfuegbare Bausteine."""
     route = resolve_route()
     audio_router = get_audio_router()
+    chatterbox_ok, openrouter_tts_ok, cartesia_ok = await _chatterbox_reachable(), await _openrouter_tts_available(), await _cartesia_available()
     return {
         "profile": active_profile(),
         "app_env": settings.app_env,
@@ -30,6 +77,9 @@ async def get_config():
             "tts_providers": sorted(TTS_REGISTRY),
             "input_endpoints": [c.model_dump() for c in await audio_router.list_inputs()],
             "output_endpoints": [c.model_dump() for c in await audio_router.list_outputs()],
+            "chatterbox_reachable": chatterbox_ok,
+            "openrouter_tts_available": openrouter_tts_ok,
+            "cartesia_available": cartesia_ok,
         },
         "secrets": {
             "openrouter_api_key_set": bool(settings.openrouter_api_key.strip()),
diff --git a/app/api/speak.py b/app/api/speak.py
index a39e8d1..e93428f 100644
--- a/app/api/speak.py
+++ b/app/api/speak.py
@@ -12,6 +12,7 @@ from app.dependencies import (
     resolve_output_endpoint,
     get_store,
     piper_voice_for_language,
+    voice_for_route,
 )
 from app.quota import enforce_quota, record_usage, QuotaExceededError
 from app.schemas import SpeakRequest
@@ -36,7 +37,7 @@ async def speak(
     try:
         route = resolve_route(user, session_id, overrides)
         # Explizit angefragte Stimme gewinnt; sonst folgt sie der Sprache (Piper-Parität zu /chat).
-        voice = payload.voice or piper_voice_for_language(route.tts_provider, route.language)
+        voice = payload.voice or voice_for_route(route.tts_provider, route.language, route.voice_gender)
         orchestrator = build_orchestrator(route)
         output = await resolve_output_endpoint(route)
     except SessionOwnershipError as exc:
diff --git a/app/api/ws.py b/app/api/ws.py
index 18d03ad..4199f5c 100644
--- a/app/api/ws.py
+++ b/app/api/ws.py
@@ -27,6 +27,7 @@ from app.dependencies import (
     build_orchestrator,
     resolve_output_endpoint,
     piper_voice_for_language,
+    voice_for_route,
 )
 from app.store import ANONYMOUS_USER_ID, SessionOwnershipError
 from app.audio.vad import EnergyVAD
@@ -122,7 +123,7 @@ async def _run_turn(
     elif effective_voice is not None:
         voice = effective_voice
     else:
-        voice = piper_voice_for_language(route.tts_provider, route.language)
+        voice = voice_for_route(route.tts_provider, route.language, route.voice_gender)
     stream = bool(options.get("stream"))
     # text_only: Geräte-TTS (Web Speech API) spricht selbst -> kein Server-Audio erzeugen/senden.
     text_only = bool(options.get("text_only"))
@@ -233,15 +234,15 @@ async def _voice_turn(websocket, store, user, session_id, audio, fmt, options):
                 audio, fmt=fmt, language=None
             )
             # Stimme folgt der erkannten Sprache (Fallback: Routensprache).
-            effective_voice = piper_voice_for_language(
-                route.tts_provider, detected_language or route.language
+            effective_voice = voice_for_route(
+                route.tts_provider, detected_language or route.language, route.voice_gender
             )
         else:
             # Fix: konfigurierte Sprache an Whisper → Whisper übersetzt automatisch.
             transcript = await orchestrator.stt.transcribe(audio, fmt=fmt, language=route.language)
             detected_language = None
             # Stimme folgt der konfigurierten Sprache.
-            effective_voice = piper_voice_for_language(route.tts_provider, route.language)
+            effective_voice = voice_for_route(route.tts_provider, route.language, route.voice_gender)
     except Exception as exc:
         await websocket.send_json({"type": "error", "status": 502, "detail": str(exc)})
         return
diff --git a/app/config.py b/app/config.py
index 14ce23f..bb2109a 100644
--- a/app/config.py
+++ b/app/config.py
@@ -141,6 +141,11 @@ class Settings(BaseSettings):
     piper_voice: str = "de_DE-thorsten-high"  # Stimmmodell-Name (ohne .onnx) oder voller Pfad
     tts_sample_rate: int = 24000              # Ziel-Sample-Rate (das Gateway erwartet 24000 Hz)
     # --- Chatterbox-TTS (hohe Qualitaet + Voice-Cloning, eigener HTTP-Dienst) -
+    # --- Cartesia TTS (Cloud, sehr niedrige Latenz) -------------------------
+    cartesia_api_key: str = ""
+    cartesia_voice_id: str = ""          # UUID aus cartesia.ai/voices (Fallback)
+    cartesia_tts_model: str = "sonic-turbo"
+    cartesia_voices: str = ""              # "lang:gender:uuid:name" je Zeile, # = Kommentar  # sonic-2 | sonic-multilingual
     chatterbox_base_url: str = "http://127.0.0.1:9999"
     chatterbox_voice: str = ""                # Pfad zu Referenz-WAV (Voice-Cloning) oder leer
     chatterbox_lang: str = "de"
diff --git a/app/dependencies.py b/app/dependencies.py
index 9179ae5..b58407d 100644
--- a/app/dependencies.py
+++ b/app/dependencies.py
@@ -18,6 +18,7 @@ from app.providers.stt.faster_whisper import FasterWhisperProvider
 from app.providers.llm.local_openai_compatible import LocalOpenAICompatibleLLM
 from app.providers.llm.openrouter import OpenRouterLLMProvider
 from app.providers.tts.openrouter import OpenRouterTTSProvider
+from app.providers.tts.cartesia import CartesiaTTSProvider
 from app.providers.tts.chatterbox import ChatterboxTTSProvider
 from app.providers.tts.piper import PiperTTSProvider
 from app.providers.fallback import (
@@ -80,6 +81,10 @@ TTS_REGISTRY = {
         s.chatterbox_timeout,
         voices_dir=s.chatterbox_voices_dir,
     ),
+    "cartesia": lambda s: CartesiaTTSProvider(
+        s.cartesia_api_key, s.cartesia_voice_id, s.cartesia_tts_model, s.tts_sample_rate,
+        voices_str=s.cartesia_voices,
+    ),
     "piper": lambda s: PiperTTSProvider(
         s.piper_bin, s.piper_voices_dir, s.piper_voice, s.tts_sample_rate
     ),
@@ -151,32 +156,74 @@ ROUTE_KEYS = (
     "tts_provider",
     "language",
     "language_mode",
+    "voice_gender",
 )
 
-# Stimm-Auswahl nach Sprache: (effektive) Sprache → beste Piper-Stimme.
-# Gilt in BEIDEN Modi — die gesprochene Stimme folgt immer der Antwortsprache.
+# Stimm-Auswahl nach Sprache: (effektive) Sprache → beste Piper-Stimme (gender-agnostisch).
 LANG_TO_PIPER_VOICE: dict[str, str] = {
     "de":  "de_DE-thorsten-high",
     "en":  "en_US-lessac-high",
     "fr":  "fr_FR-siwis-medium",
     "es":  "es_ES-sharvard-medium",
     "it":  "it_IT-paola-medium",
-    "nl":  "nl_NL-mls-medium",
+    "pt":  "pt_BR-faber-medium",
+    "pl":  "pl_PL-darkman-medium",
+    "ar":  "ar_JO-kareem-medium",
     "ru":  "ru_RU-irina-medium",
     "zh":  "zh_CN-huayan-medium",
     "cmn": "zh_CN-huayan-medium",
 }
 
+# Geschlechtsspezifische Piper-Stimmen für Sprachen mit m+f Varianten.
+# Präzedenz: PIPER_VOICE_GENDERED > LANG_TO_PIPER_VOICE.
+# ES sharvard ist ein Multi-Speaker-Modell: #0=M, #1=F (speaker_id_map: {'M':0,'F':1})
+PIPER_VOICE_GENDERED: dict[tuple[str, str], str] = {
+    ("de", "f"): "de_DE-kerstin-low",
+    ("de", "m"): "de_DE-thorsten-high",
+    ("en", "f"): "en_US-amy-medium",
+    ("en", "m"): "en_US-lessac-high",
+    ("fr", "f"): "fr_FR-siwis-medium",
+    ("fr", "m"): "fr_FR-tom-medium",
+    ("es", "f"): "es_ES-sharvard-medium#1",
+    ("es", "m"): "es_ES-sharvard-medium#0",
+    ("it", "f"): "it_IT-paola-medium",
+    ("it", "m"): "it_IT-riccardo-x_low",
+    ("ru", "f"): "ru_RU-irina-medium",
+    ("ru", "m"): "ru_RU-ruslan-medium",
+    ("pl", "f"): "pl_PL-gosia-medium",
+    ("pl", "m"): "pl_PL-darkman-medium",
+}
+
 
 def piper_voice_for_language(tts_provider: str, language: str | None) -> str | None:
-    """Liefert die zur Sprache passende Piper-Stimme (sonst None → Provider-Default).
+    """Kompatibilitäts-Alias für voice_for_route ohne gender."""
+    return voice_for_route(tts_provider, language, "any")
 
-    Nur für Piper sinnvoll: Cloud-TTS (OpenRouter) und Chatterbox haben eigene
-    Stimmnamen (z. B. "Zephyr") und steuern die Sprache anders. Bei None nimmt der
-    jeweilige Provider seine konfigurierte Default-Stimme.
+
+def voice_for_route(
+    tts_provider: str, language: str | None, voice_gender: str = "any"
+) -> str | None:
+    """Liefert den voice-Parameter für synthesize() passend zum Provider.
+
+    - piper:    Piper-Stimmmodell-Name; geschlechtsspezifisch wenn Varianten vorhanden
+    - cartesia: Geschlechts-Code ("m"/"f"/"any") → CartesiaTTSProvider löst UUID auf
+    - andere:   None (Provider nutzt seinen eigenen Default)
     """
     if tts_provider == "piper" and language:
-        return LANG_TO_PIPER_VOICE.get(language)
+        lang = language.lower()
+        gender = (voice_gender or "any").lower()
+        # Geschlechtsspezifische Stimme: bevorzugte Richtung, dann Gegenteil
+        if gender in ("m", "f"):
+            candidates = [gender, "f" if gender == "m" else "m"]
+        else:  # "any" → weiblich bevorzugt (konsistent mit Cartesia)
+            candidates = ["f", "m"]
+        for g in candidates:
+            voice = PIPER_VOICE_GENDERED.get((lang, g))
+            if voice:
+                return voice
+        return LANG_TO_PIPER_VOICE.get(lang)
+    if tts_provider == "cartesia":
+        return voice_gender or "any"
     return None
 
 
@@ -189,6 +236,7 @@ class ResolvedRoute:
     tts_provider: str
     language: str
     language_mode: str = "fix"
+    voice_gender: str = "any"
 
     def as_dict(self) -> dict:
         return {
@@ -199,6 +247,7 @@ class ResolvedRoute:
             "tts_provider": self.tts_provider,
             "language": self.language,
             "language_mode": self.language_mode,
+            "voice_gender": self.voice_gender,
         }
 
 
diff --git a/app/providers/llm/openrouter.py b/app/providers/llm/openrouter.py
index 408afa7..f29864a 100644
--- a/app/providers/llm/openrouter.py
+++ b/app/providers/llm/openrouter.py
@@ -1,3 +1,5 @@
+import asyncio
+import time
 from collections.abc import AsyncIterator
 
 import httpx
@@ -9,6 +11,37 @@ from app.providers.llm.base import (
     sse_delta,
 )
 
+_RETRY_STATUS = {404, 429, 500, 502, 503}
+
+
+def _retry_wait(status: int, headers: httpx.Headers, attempt: int, body: str = "") -> float:
+    """Wartezeit: JSON retry_after_seconds > Retry-After > X-RateLimit-Reset > exponentiell."""
+    if status == 429:
+        if body:
+            try:
+                import json as _json
+                d = _json.loads(body)
+                secs = d.get("error", {}).get("metadata", {}).get("retry_after_seconds")
+                if secs:
+                    return max(1.0, min(float(secs) + 0.5, 120.0))
+            except Exception:
+                pass
+        retry_after = headers.get("Retry-After")
+        if retry_after:
+            try:
+                return max(1.0, min(float(retry_after) + 0.5, 120.0))
+            except ValueError:
+                pass
+        reset_ms = headers.get("X-RateLimit-Reset")
+        if reset_ms:
+            try:
+                wait = (int(reset_ms) / 1000.0) - time.time() + 0.5
+                return max(1.0, min(wait, 120.0))
+            except ValueError:
+                pass
+        return 30.0
+    return 2.0 * attempt
+
 
 SYSTEM_PROMPT = """
 You are a voice assistant for spoken conversations with older adults.
@@ -52,9 +85,10 @@ Always optimize your answer for listening, not for reading.
 
 
 class OpenRouterLLMProvider(LLMProvider):
-    def __init__(self, api_key: str, model: str):
+    def __init__(self, api_key: str, model: str, max_retries: int = 5):
         self.api_key = (api_key or "").strip()
         self.model = (model or "").strip()
+        self.max_retries = max(1, max_retries)
 
     def _build_messages(
         self, text: str, history: list[dict] | None, language: str | None = None
@@ -74,7 +108,6 @@ class OpenRouterLLMProvider(LLMProvider):
         messages = [{"role": "system", "content": system_content}]
         if history:
             messages.extend(history)
-        # Sprach-Erinnerung direkt an der letzten Nutzer-Nachricht (schlägt History-Trägheit).
         messages.append({"role": "user", "content": with_lang_reminder(text.strip(), language)})
         return messages
 
@@ -89,40 +122,52 @@ class OpenRouterLLMProvider(LLMProvider):
             "model": self.model,
             "messages": self._build_messages(text, history, language=language),
         }
-
         timeout = httpx.Timeout(connect=10.0, read=120.0, write=30.0, pool=10.0)
+        last_error: Exception | None = None
 
-        async with httpx.AsyncClient(timeout=timeout) as client:
+        for attempt in range(1, self.max_retries + 1):
+            async with httpx.AsyncClient(timeout=timeout) as client:
+                try:
+                    response = await client.post(
+                        "https://openrouter.ai/api/v1/chat/completions",
+                        headers={
+                            "Authorization": f"Bearer {self.api_key}",
+                            "Content-Type": "application/json",
+                        },
+                        json=payload,
+                    )
+                    if response.status_code in _RETRY_STATUS and attempt < self.max_retries:
+                        last_error = RuntimeError(
+                            f"OpenRouter LLM error {response.status_code}: {response.text}"
+                        )
+                        await asyncio.sleep(_retry_wait(response.status_code, response.headers, attempt, response.text))
+                        continue
+                    response.raise_for_status()
+                except httpx.HTTPStatusError as exc:
+                    last_error = RuntimeError(
+                        f"OpenRouter LLM error {exc.response.status_code}: {exc.response.text}"
+                    )
+                    if exc.response.status_code in _RETRY_STATUS and attempt < self.max_retries:
+                        await asyncio.sleep(_retry_wait(exc.response.status_code, exc.response.headers, attempt, exc.response.text))
+                        continue
+                    raise last_error from exc
+                except httpx.TimeoutException as exc:
+                    raise RuntimeError("OpenRouter LLM timeout") from exc
+                except httpx.HTTPError as exc:
+                    raise RuntimeError(f"OpenRouter LLM transport error: {exc}") from exc
+
+            data = response.json()
             try:
-                response = await client.post(
-                    "https://openrouter.ai/api/v1/chat/completions",
-                    headers={
-                        "Authorization": f"Bearer {self.api_key}",
-                        "Content-Type": "application/json",
-                    },
-                    json=payload,
-                )
-                response.raise_for_status()
-            except httpx.HTTPStatusError as exc:
-                raise RuntimeError(
-                    f"OpenRouter LLM error {exc.response.status_code}: {exc.response.text}"
-                ) from exc
-            except httpx.TimeoutException as exc:
-                raise RuntimeError("OpenRouter LLM timeout") from exc
-            except httpx.HTTPError as exc:
-                raise RuntimeError(f"OpenRouter LLM transport error: {exc}") from exc
+                content = data["choices"][0]["message"]["content"]
+            except (KeyError, IndexError, TypeError) as exc:
+                raise RuntimeError(f"Unexpected OpenRouter LLM response: {data}") from exc
 
-        data = response.json()
+            if not content or not str(content).strip():
+                raise RuntimeError("OpenRouter LLM returned empty content")
 
-        try:
-            content = data["choices"][0]["message"]["content"]
-        except (KeyError, IndexError, TypeError) as exc:
-            raise RuntimeError(f"Unexpected OpenRouter LLM response: {data}") from exc
+            return str(content).strip()
 
-        if not content or not str(content).strip():
-            raise RuntimeError("OpenRouter LLM returned empty content")
-
-        return str(content).strip()
+        raise last_error or RuntimeError("OpenRouter LLM: all attempts failed")
 
     async def stream(
         self,
@@ -137,30 +182,42 @@ class OpenRouterLLMProvider(LLMProvider):
             "stream": True,
         }
         timeout = httpx.Timeout(connect=10.0, read=120.0, write=30.0, pool=10.0)
+        last_error: Exception | None = None
 
-        async with httpx.AsyncClient(timeout=timeout) as client:
-            try:
-                async with client.stream(
-                    "POST",
-                    "https://openrouter.ai/api/v1/chat/completions",
-                    headers={
-                        "Authorization": f"Bearer {self.api_key}",
-                        "Content-Type": "application/json",
-                    },
-                    json=payload,
-                ) as response:
-                    if response.status_code >= 400:
-                        body = await response.aread()
-                        raise RuntimeError(
-                            f"OpenRouter LLM error {response.status_code}: "
-                            f"{body.decode(errors='replace')}"
-                        )
-                    async for line in response.aiter_lines():
-                        delta = sse_delta(line)
-                        if delta:
-                            yield delta
-            except httpx.TimeoutException as exc:
-                raise RuntimeError("OpenRouter LLM timeout") from exc
-            except httpx.HTTPError as exc:
-                raise RuntimeError(f"OpenRouter LLM transport error: {exc}") from exc
+        for attempt in range(1, self.max_retries + 1):
+            async with httpx.AsyncClient(timeout=timeout) as client:
+                try:
+                    async with client.stream(
+                        "POST",
+                        "https://openrouter.ai/api/v1/chat/completions",
+                        headers={
+                            "Authorization": f"Bearer {self.api_key}",
+                            "Content-Type": "application/json",
+                        },
+                        json=payload,
+                    ) as response:
+                        if response.status_code in _RETRY_STATUS and attempt < self.max_retries:
+                            body = await response.aread()
+                            last_error = RuntimeError(
+                                f"OpenRouter LLM error {response.status_code}: "
+                                f"{body.decode(errors='replace')}"
+                            )
+                            await asyncio.sleep(_retry_wait(response.status_code, response.headers, attempt, body.decode(errors='replace')))
+                            continue
+                        if response.status_code >= 400:
+                            body = await response.aread()
+                            raise RuntimeError(
+                                f"OpenRouter LLM error {response.status_code}: "
+                                f"{body.decode(errors='replace')}"
+                            )
+                        async for line in response.aiter_lines():
+                            delta = sse_delta(line)
+                            if delta:
+                                yield delta
+                        return
+                except httpx.TimeoutException as exc:
+                    raise RuntimeError("OpenRouter LLM timeout") from exc
+                except httpx.HTTPError as exc:
+                    raise RuntimeError(f"OpenRouter LLM transport error: {exc}") from exc
 
+        raise last_error or RuntimeError("OpenRouter LLM: all attempts failed")
diff --git a/app/providers/tts/cartesia.py b/app/providers/tts/cartesia.py
new file mode 100644
index 0000000..fbcbe21
--- /dev/null
+++ b/app/providers/tts/cartesia.py
@@ -0,0 +1,142 @@
+import httpx
+
+from app.providers.tts.base import TTSProvider
+
+_API_URL = "https://api.cartesia.ai/tts/bytes"
+_API_VERSION = "2024-06-10"
+
+# Sprachen, die sonic-turbo unterstützt (Stand 2026-06).
+_LANG_MAP = {
+    "de": "de", "en": "en", "fr": "fr", "es": "es",
+    "zh": "zh", "cmn": "zh", "pt": "pt", "ja": "ja", "ko": "ko", "hi": "hi",
+}
+
+# sonic-turbo unterstützt diese Sprachen nicht — Fallback auf Piper empfohlen.
+_UNSUPPORTED_LANGS = {"ar", "it", "ru", "nl", "pl", "sv", "tr"}
+
+_LANG_NAMES = {
+    "ar": "Arabisch", "it": "Italienisch", "ru": "Russisch",
+    "nl": "Niederländisch", "pl": "Polnisch", "sv": "Schwedisch", "tr": "Türkisch",
+}
+
+_GENDER_CODES = {"m", "f", "any"}
+
+
+def _parse_voice_map(voices_str: str) -> dict[tuple[str, str], dict]:
+    """Parst 'lang:gender:uuid:name'-Zeilen in ein (lang,gender)→{uuid,name}-Dict."""
+    result = {}
+    for line in (voices_str or "").splitlines():
+        line = line.strip()
+        if not line or line.startswith("#"):
+            continue
+        parts = line.split(":", 3)
+        if len(parts) != 4:
+            continue
+        lang, gender, uuid, name = [p.strip() for p in parts]
+        if lang and gender in ("m", "f") and uuid:
+            result[(lang, gender)] = {"uuid": uuid, "name": name}
+    return result
+
+
+class CartesiaTTSProvider(TTSProvider):
+    def __init__(
+        self,
+        api_key: str,
+        voice_id: str,
+        model: str = "sonic-2",
+        sample_rate: int = 24000,
+        voices_str: str = "",
+    ):
+        self.api_key = (api_key or "").strip()
+        self.default_voice_id = (voice_id or "").strip()
+        self.model = (model or "sonic-2").strip()
+        self.sample_rate = int(sample_rate)
+        self.voice_map = _parse_voice_map(voices_str)
+
+    def _resolve_voice_id(self, language: str | None, gender_hint: str | None) -> str:
+        """Wählt UUID aus Voice-Map. Fallback-Kette: f→m→any→default_voice_id."""
+        lang = (language or "de").lower()
+        gender = (gender_hint or "any").strip().lower()
+
+        # Präferenz-Reihenfolge: "any" → weiblich bevorzugt
+        if gender == "any" or gender not in ("m", "f"):
+            candidates = ["f", "m"]
+        elif gender == "f":
+            candidates = ["f", "m"]
+        else:
+            candidates = ["m", "f"]
+
+        for g in candidates:
+            entry = self.voice_map.get((lang, g))
+            if entry:
+                return entry["uuid"]
+
+        return self.default_voice_id
+
+    async def synthesize(
+        self,
+        text: str,
+        voice: str | None = None,
+        audio_format: str = "pcm",
+        language: str | None = None,
+    ) -> bytes:
+        if not self.api_key:
+            raise ValueError("CARTESIA_API_KEY is empty")
+        if not text or not text.strip():
+            raise ValueError("TTS input text is empty")
+
+        # voice ist entweder ein Geschlechts-Code ("m"/"f"/"any") oder eine direkte UUID
+        if voice and voice not in _GENDER_CODES:
+            voice_id = voice  # direkte UUID-Übergabe
+        else:
+            voice_id = self._resolve_voice_id(language, voice)
+
+        if not voice_id:
+            raise ValueError("CARTESIA_VOICE_ID ist nicht konfiguriert — bitte UUID in voice-assistant.toml eintragen")
+
+        lang_key = (language or "de").lower()
+        if lang_key in _UNSUPPORTED_LANGS:
+            lang_label = _LANG_NAMES.get(lang_key, lang_key)
+            raise RuntimeError(
+                f"Cartesia unterstützt {lang_label} ({lang_key.upper()}) noch nicht als TTS-Sprache"
+                " — bitte Piper oder einen anderen Provider wählen"
+            )
+        lang_code = _LANG_MAP.get(lang_key, "de")
+
+        payload = {
+            "model_id": self.model,
+            "transcript": text.strip(),
+            "voice": {"mode": "id", "id": voice_id},
+            "output_format": {
+                "container": "raw",
+                "encoding": "pcm_s16le",
+                "sample_rate": self.sample_rate,
+            },
+            "language": lang_code,
+        }
+
+        timeout = httpx.Timeout(connect=10.0, read=60.0, write=10.0, pool=10.0)
+        async with httpx.AsyncClient(timeout=timeout) as client:
+            try:
+                response = await client.post(
+                    _API_URL,
+                    headers={
+                        "X-API-Key": self.api_key,
+                        "Cartesia-Version": _API_VERSION,
+                        "Content-Type": "application/json",
+                    },
+                    json=payload,
+                )
+                response.raise_for_status()
+            except httpx.HTTPStatusError as exc:
+                raise RuntimeError(
+                    f"Cartesia TTS error {exc.response.status_code}: {exc.response.text}"
+                ) from exc
+            except httpx.TimeoutException as exc:
+                raise RuntimeError("Cartesia TTS timeout") from exc
+            except httpx.HTTPError as exc:
+                raise RuntimeError(f"Cartesia TTS transport error: {exc}") from exc
+
+        if not response.content:
+            raise RuntimeError("Cartesia TTS returned empty audio")
+        return response.content
diff --git a/app/providers/tts/piper.py b/app/providers/tts/piper.py
index e1d102c..f19cd04 100644
--- a/app/providers/tts/piper.py
+++ b/app/providers/tts/piper.py
@@ -104,6 +104,17 @@ class PiperTTSProvider(TTSProvider):
         except (OSError, KeyError, ValueError, TypeError):
             return self.target_rate  # Konfig unlesbar -> Resampling überspringen
 
+    @staticmethod
+    def _parse_speaker(voice: str | None) -> tuple[str | None, int | None]:
+        """Trennt 'stimme#speaker_id' auf. Liefert (stimme, int_id) oder (stimme, None)."""
+        if voice and "#" in voice:
+            name, sid = voice.rsplit("#", 1)
+            try:
+                return name, int(sid)
+            except ValueError:
+                pass
+        return voice, None
+
     async def synthesize(
         self,
         text: str,
@@ -114,9 +125,10 @@ class PiperTTSProvider(TTSProvider):
         if not text or not text.strip():
             raise ValueError("TTS input text is empty")
 
-        model, config = self._model_paths(voice)
+        voice_name, speaker_id = self._parse_speaker(voice)
+        model, config = self._model_paths(voice_name)
         pcm, native_rate = await asyncio.to_thread(
-            self._synthesize_sync, str(model), str(config), text.strip()
+            self._synthesize_sync, str(model), str(config), text.strip(), speaker_id
         )
         if native_rate != self.target_rate:
             pcm = await asyncio.to_thread(_resample, pcm, native_rate, self.target_rate)
@@ -125,26 +137,30 @@ class PiperTTSProvider(TTSProvider):
             return _wrap_wav(pcm, self.target_rate)
         return pcm
 
-    def _synthesize_sync(self, model: str, config: str, text: str) -> tuple[bytes, int]:
+    def _synthesize_sync(self, model: str, config: str, text: str, speaker_id: int | None = None) -> tuple[bytes, int]:
         if _PIPER_LIB:
-            voice = _load_voice(model)
+            from piper.config import SynthesisConfig
+            piper_voice = _load_voice(model)
+            syn_cfg = SynthesisConfig(speaker_id=speaker_id) if speaker_id is not None else None
             pcm = bytearray()
             rate = self.target_rate
-            for chunk in voice.synthesize(text):
+            for chunk in piper_voice.synthesize(text, syn_config=syn_cfg):
                 pcm += chunk.audio_int16_bytes
                 rate = chunk.sample_rate
             if not pcm:
                 raise RuntimeError("Piper lieferte kein Audio")
             return bytes(pcm), rate
         # Fallback: piper-Binary (Subprozess pro Aufruf, langsamer).
-        return self._run_piper_binary(model, config, text)
+        return self._run_piper_binary(model, config, text, speaker_id)
 
-    def _run_piper_binary(self, model: str, config: str, text: str) -> tuple[bytes, int]:
+    def _run_piper_binary(self, model: str, config: str, text: str, speaker_id: int | None = None) -> tuple[bytes, int]:
         import subprocess
 
         if not (shutil.which(self.bin_path) or Path(self.bin_path).exists()):
             raise RuntimeError(f"Piper-Binary nicht gefunden: {self.bin_path}")
         cmd = [self.bin_path, "-m", model, "-c", config, "--output-raw"]
+        if speaker_id is not None:
+            cmd += ["--speaker", str(speaker_id)]
         proc = subprocess.run(cmd, input=text.encode("utf-8"), capture_output=True)
         if proc.returncode != 0:
             detail = proc.stderr.decode("utf-8", "replace").strip()[:300]
diff --git a/app/web/app.js b/app/web/app.js
index a7ce676..234caa9 100644
--- a/app/web/app.js
+++ b/app/web/app.js
@@ -20,6 +20,29 @@ function langOverrides() {
 }
 
 // Nutzer-Präferenzen als Request-Overrides für jeden Turn.
+// Geschlechtspräferenz: 'm', 'f', oder 'any' (= weiblich bevorzugt)
+let _voiceGender = 'f';
+
+function applyGenderChoice(gender) {
+  // 'any' aus alten Prefs → weiblich als Standardwert
+  _voiceGender = (gender === 'm') ? 'm' : 'f';
+  const btnM = document.getElementById('gender-m');
+  const btnF = document.getElementById('gender-f');
+  if (!btnM || !btnF) return;
+  const mActive = (_voiceGender === 'm');
+  const fActive = (_voiceGender === 'f');
+  // Aktiv: blauer Hintergrund + weiße Schrift
+  // Inaktiv: kein Hintergrund + dunkle Schrift (explizit, damit kein Vererbungsproblem)
+  btnM.classList.toggle('bg-blue-600', mActive);
+  btnM.classList.toggle('text-white', mActive);
+  btnM.classList.toggle('border-blue-600', mActive);
+  btnM.classList.toggle('text-slate-700', !mActive);
+  btnF.classList.toggle('bg-blue-600', fActive);
+  btnF.classList.toggle('text-white', fActive);
+  btnF.classList.toggle('border-blue-600', fActive);
+  btnF.classList.toggle('text-slate-700', !fActive);
+}
+
 function overrides() {
   const out = { ...langOverrides() };
   if (isDeviceMode()) {
@@ -27,6 +50,7 @@ function overrides() {
   } else if (ttsSel && ttsSel.value) {
     out.tts_provider = ttsSel.value;
   }
+  out.voice_gender = _voiceGender;
   return out;
 }
 
@@ -45,7 +69,7 @@ async function savePrefs(obj) {
 const TTS_SUPPORTED = typeof window !== "undefined" && "speechSynthesis" in window;
 const LANG_BCP47 = {
   de: "de-DE", en: "en-US", fr: "fr-FR", es: "es-ES",
-  it: "it-IT", nl: "nl-NL", ru: "ru-RU", zh: "zh-CN",
+  it: "it-IT", pt: "pt-BR", pl: "pl-PL", ar: "ar-SA", ru: "ru-RU", zh: "zh-CN",
 };
 
 function isMobile() {
@@ -179,6 +203,69 @@ function refreshDevicePreset() {
   syncPresetUI();
 }
 
+// Graut "Hohe Qualität" (Chatterbox) aus, wenn kein GPU-Backend erreichbar ist.
+async function refreshChatterboxPreset() {
+  const btn = document.querySelector('.tts-preset[data-tts="chatterbox"]');
+  if (!btn || !ttsSel) return;
+  try {
+    const cfg = await fetch("/api/config", { cache: "no-store" }).then(r => r.json());
+    const available = cfg.available?.chatterbox_reachable === true;
+    btn.disabled = !available;
+    btn.classList.toggle("opacity-50", !available);
+    btn.classList.toggle("cursor-not-allowed", !available);
+    const sub = btn.querySelector(".tts-sub");
+    if (sub) sub.textContent = available
+      ? "Chatterbox · natürliche Stimme"
+      : "Chatterbox · kein GPU-Backend erreichbar";
+    if (!available && ttsSel.value === "chatterbox") {
+      ttsSel.value = "piper";
+      syncPresetUI();
+    }
+  } catch (e) { /* ignorieren */ }
+}
+
+// Graut "Cartesia" aus, wenn API-Key oder Voice-ID fehlt.
+async function refreshCartesiaPreset() {
+  const btn = document.querySelector('.tts-preset[data-tts="cartesia"]');
+  if (!btn || !ttsSel) return;
+  try {
+    const cfg = await fetch("/api/config", { cache: "no-store" }).then(r => r.json());
+    const available = cfg.available?.cartesia_available === true;
+    btn.disabled = !available;
+    btn.classList.toggle("opacity-50", !available);
+    btn.classList.toggle("cursor-not-allowed", !available);
+    const sub = btn.querySelector(".tts-sub");
+    if (sub) sub.textContent = available
+      ? "Cartesia Sonic · natürlich & schnell"
+      : "Cartesia · kein API-Key konfiguriert";
+    if (!available && ttsSel.value === "cartesia") {
+      ttsSel.value = "piper";
+      syncPresetUI();
+    }
+  } catch (e) { /* ignorieren */ }
+}
+
+// Graut "Cloud" (OpenRouter TTS) aus, wenn das TTS-Modell nicht verfügbar ist.
+async function refreshOpenrouterTtsPreset() {
+  const btn = document.querySelector('.tts-preset[data-tts="openrouter"]');
+  if (!btn || !ttsSel) return;
+  try {
+    const cfg = await fetch("/api/config", { cache: "no-store" }).then(r => r.json());
+    const available = cfg.available?.openrouter_tts_available === true;
+    btn.disabled = !available;
+    btn.classList.toggle("opacity-50", !available);
+    btn.classList.toggle("cursor-not-allowed", !available);
+    const sub = btn.querySelector(".tts-sub");
+    if (sub) sub.textContent = available
+      ? "OpenRouter · benötigt Internet"
+      : "OpenRouter · kein TTS-Modell verfügbar";
+    if (!available && ttsSel.value === "openrouter") {
+      ttsSel.value = "piper";
+      syncPresetUI();
+    }
+  } catch (e) { /* ignorieren */ }
+}
+
 // Markiert das aktive Ton-Preset (Häkchen + Rahmen) passend zu #tts.
 function syncPresetUI() {
   const val = ttsSel ? ttsSel.value : "";
@@ -241,9 +328,10 @@ function loadMe() {
         // Flex-Modus -> "flex", sonst die feste Sprache (Fallback: Flex).
         langSel.value = prefs.language_mode === "flex"
           ? "flex"
-          : (prefs.language || "flex");
+          : (prefs.language || "de");
       }
       applyPlaybackChoice(prefs.tts_provider);
+      applyGenderChoice(prefs.voice_gender || 'f');
       return me;
     } catch (e) {
       $("#identity").textContent = "Verbindung fehlgeschlagen";
@@ -291,6 +379,7 @@ function addMessage(role, text, meta = {}) {
   div.className = BUBBLE[role] || BUBBLE.assistant;
   const span = document.createElement("span");
   span.className = "bubble-text";
+  span.dir = "auto";
   span.textContent = text;
   div.appendChild(span);
   div._textEl = span;
@@ -684,7 +773,25 @@ if (ttsSel) {
   });
 }
 
+
+// Gender-Buttons
+document.addEventListener("DOMContentLoaded", () => {
+  const btnM = document.getElementById("gender-m");
+  const btnF = document.getElementById("gender-f");
+  if (btnM) btnM.addEventListener("click", () => {
+    applyGenderChoice("m");
+    savePrefs({ voice_gender: "m" });
+  });
+  if (btnF) btnF.addEventListener("click", () => {
+    applyGenderChoice("f");
+    savePrefs({ voice_gender: "f" });
+  });
+});
+
 loadMe();
+refreshChatterboxPreset();
+refreshCartesiaPreset();
+refreshOpenrouterTtsPreset();
 
 // ═══════════════════════════════════════════════════════════════
 // ADMIN PANEL
@@ -1570,12 +1677,12 @@ const FIELD_META = {
   default_stt_provider:            { ui: "combo",   opts: ["openrouter","faster-whisper"], test: "stt" },
   default_llm_provider:            { ui: "combo",   opts: ["openrouter","local-openai-compatible"], test: "llm" },
   default_tts_provider:            { ui: "combo",   opts: ["openrouter","piper","chatterbox"], test: "tts" },
-  default_language:                { ui: "combo",   opts: ["de","en","fr","es","it","nl"],
-                                     labels: { de:"Deutsch",en:"Englisch",fr:"Französisch",es:"Spanisch",it:"Italienisch",nl:"Niederländisch" } },
+  default_language:                { ui: "combo",   opts: ["de","en","fr","es","it","pt","pl","ar"],
+                                     labels: { de:"Deutsch",en:"Englisch",fr:"Französisch",es:"Spanisch",it:"Italienisch",pt:"Portugiesisch",pl:"Polnisch",ar:"Arabisch" } },
   openrouter_llm_model:            { ui: "combo",   opts: ["google/gemini-3.1-flash-lite","openai/gpt-4.1-mini","google/gemini-2.5-flash","anthropic/claude-haiku-4-5-20251001","meta-llama/llama-3.3-70b-instruct"], test: "llm" },
   openrouter_tts_model:            { ui: "combo",   opts: ["google/gemini-3.1-flash-tts-preview","openai/gpt-4o-mini-tts"], test: "tts", testProvider: "openrouter" },
   openrouter_tts_voice:            { ui: "combo",   opts: ["Zephyr","Puck","Charon","Kore","Fenrir","Leda","Orus","Aoede","Callirrhoe","Enceladus","Algieba","Despina","Achernar"], test: "tts", testProvider: "openrouter" },
-  piper_voice:                     { ui: "combo",   opts: ["de_DE-thorsten-high","en_US-ryan-high","en_US-lessac-high","en_GB-cori-high","es_ES-sharvard-medium","fr_FR-siwis-medium","it_IT-paola-medium","nl_NL-mls-medium","ru_RU-irina-medium","zh_CN-huayan-medium"], test: "tts", testProvider: "piper" },
+  piper_voice:                     { ui: "combo",   opts: ["de_DE-thorsten-high","de_DE-kerstin-low","en_US-lessac-high","en_US-amy-medium","en_US-ryan-high","en_GB-cori-high","fr_FR-siwis-medium","fr_FR-tom-medium","es_ES-sharvard-medium","it_IT-paola-medium","it_IT-riccardo-x_low","pt_BR-faber-medium","pl_PL-gosia-medium","pl_PL-darkman-medium","ar_JO-kareem-medium","ru_RU-irina-medium","ru_RU-ruslan-medium","zh_CN-huayan-medium"], test: "tts", testProvider: "piper" },
   local_llm_system_prompt:         { ui: "textarea", test: "llm" },
   local_llm_temperature:           { ui: "range",   min: 0, max: 2, step: 0.1, test: "llm" },
   local_llm_top_p:                 { ui: "range",   min: 0, max: 1, step: 0.05, test: "llm" },
diff --git a/app/web/index.html b/app/web/index.html
index fafb5b0..ea91fc4 100644
--- a/app/web/index.html
+++ b/app/web/index.html
@@ -163,7 +163,9 @@
                 <option value="fr">Französisch (fr)</option>
                 <option value="es">Spanisch (es)</option>
                 <option value="it">Italienisch (it)</option>
-                <option value="nl">Niederländisch (nl)</option>
+                <option value="pt">Portugiesisch (pt)</option>
+                <option value="pl">Polnisch (pl)</option>
+                <option value="ar">Arabisch (ar)</option>
                 <option value="ru">Russisch (ru)</option>
                 <option value="zh">Chinesisch (zh)</option>
               </select>
@@ -225,7 +227,9 @@
       <option value="fr">🇫🇷 FR</option>
       <option value="es">🇪🇸 ES</option>
       <option value="it">🇮🇹 IT</option>
-      <option value="nl">🇳🇱 NL</option>
+      <option value="pt">🇧🇷 PT</option>
+      <option value="pl">🇵🇱 PL</option>
+      <option value="ar">🇸🇦 AR</option>
       <option value="ru">🇷🇺 RU</option>
       <option value="zh">🇨🇳 ZH</option>
     </select>
@@ -244,6 +248,7 @@
     <option value="device">Im Gerät</option>
     <option value="piper">Server</option>
     <option value="chatterbox">Beste Qualität</option>
+    <option value="cartesia">Cartesia</option>
     <option value="openrouter">Cloud</option>
   </select>
 
@@ -278,6 +283,13 @@
             <span class="block text-xs text-slate-500 dark:text-slate-400">Chatterbox · natürliche Stimme</span></span>
           <span class="check text-blue-600 dark:text-blue-400 font-bold opacity-0">✓</span>
         </button>
+        <button type="button" data-tts="cartesia"
+          class="tts-preset w-full flex items-center gap-3 rounded-lg border border-slate-200 dark:border-slate-700 px-3 py-2 text-left hover:bg-slate-50 dark:hover:bg-slate-700/50 transition-colors">
+          <span class="text-lg shrink-0">🎙</span>
+          <span class="flex-1 min-w-0"><span class="block text-sm font-medium">Cartesia</span>
+            <span class="tts-sub block text-xs text-slate-500 dark:text-slate-400">Cartesia Sonic · natürlich &amp; schnell</span></span>
+          <span class="check text-blue-600 dark:text-blue-400 font-bold opacity-0">✓</span>
+        </button>
         <button type="button" data-tts="openrouter"
           class="tts-preset w-full flex items-center gap-3 rounded-lg border border-slate-200 dark:border-slate-700 px-3 py-2 text-left hover:bg-slate-50 dark:hover:bg-slate-700/50 transition-colors">
           <span class="text-lg shrink-0">☁</span>
@@ -288,6 +300,22 @@
       </div>
     </div>
 
+
+    <!-- Geschlechtspräferenz (wirkt bei Cartesia) -->
+    <div id="gender-pref-row">
+      <p class="text-xs font-semibold uppercase tracking-wide text-slate-500 dark:text-slate-400 mb-1.5 px-1">Stimme</p>
+      <div class="flex gap-2">
+        <button type="button" id="gender-m"
+          class="gender-btn flex-1 rounded-lg border border-slate-200 dark:border-slate-700 py-1.5 text-sm font-medium text-slate-700 transition-colors hover:bg-slate-50 dark:hover:bg-slate-700/50">
+          ♂ Männlich
+        </button>
+        <button type="button" id="gender-f"
+          class="gender-btn flex-1 rounded-lg border border-slate-200 dark:border-slate-700 py-1.5 text-sm font-medium text-slate-700 transition-colors hover:bg-slate-50 dark:hover:bg-slate-700/50">
+          ♀ Weiblich
+        </button>
+      </div>
+    </div>
+
     <!-- Aktionen -->
     <div class="border-t border-slate-100 dark:border-slate-700 pt-2 space-y-1">
       <button id="new-chat" type="button"
@@ -314,7 +342,7 @@
     <form id="prompt-form" class="flex items-center gap-2 w-full max-w-3xl mx-auto">
       <button type="button" id="mic" title="Mikrofon" aria-label="Mikrofon"
         class="shrink-0 h-11 w-11 grid place-items-center rounded-full bg-emerald-600 text-white text-lg">🎤</button>
-      <input id="prompt" type="text" autocomplete="off" placeholder="Nachricht …"
+      <input id="prompt" type="text" autocomplete="off" dir="auto" placeholder="Nachricht …"
         class="flex-1 min-w-0 rounded-xl border border-slate-300 dark:border-slate-600 bg-transparent px-4 py-2.5 focus:outline-none focus:ring-2 focus:ring-blue-500" />
       <button type="submit" id="send" title="Senden" aria-label="Senden"
         class="shrink-0 h-11 w-11 grid place-items-center rounded-full bg-blue-600 hover:bg-blue-700 text-white transition-colors">