infinicode 2.8.26 → 2.8.28

package/packages/robopark/scheduler/main.py

@@ -83,6 +83,7 @@ class Session(BaseModel):
     ended_at: Optional[datetime] = None
     end_reason: Optional[str] = None
     duration_seconds: Optional[int] = None
+    voice_config: Optional[dict] = None
 
 class GpuModel(BaseModel):
     name: str
@@ -183,6 +184,55 @@ class LiveKitTokenResponse(BaseModel):
 # DATABASE
 # =============================================================================
 
+async def _seed_voice_stacks_and_characters(db):
+    """Seed default voice stacks and production park character presets.
+
+    Idempotent: uses INSERT OR IGNORE so existing rows are never overwritten.
+    This lets operators customize via the dashboard without losing changes on
+    scheduler restart."""
+    default_stacks = [
+        (
+            "english-default", "English Default",
+            "speaches", "Systran/faster-whisper-small", "en",
+            "ollama", "gemma4:e4b",
+            "elevenlabs", "21m00Tcm4TlvDq8ikWAM", "en",
+            1, 0.7, 20, 0,
+        ),
+        (
+            "hebrew-default", "Hebrew Default",
+            "speaches", "ivrit-ai/whisper-large-v3-turbo-ct2", "he",
+            "ollama", "gemma4:e4b",
+            "elevenlabs", "21m00Tcm4TlvDq8ikWAM", "he",
+            1, 0.7, 20, 0,
+        ),
+    ]
+    for row in default_stacks:
+        await db.execute(
+            """INSERT OR IGNORE INTO voice_stacks
+               (id, name, stt_provider, stt_model, stt_language,
+                llm_provider, llm_model, tts_provider, tts_voice, tts_language,
+                allow_interruptions, min_endpointing_delay, max_turns, wake_word_enabled)
+               VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
+            row,
+        )
+
+    presets = [
+        ("volt", "VOLT", "A fast, electric-blue sports robot. Loves speed and racing.", "english-default", _json.dumps(["drive", "boost"])),
+        ("tesla", "TESLA", "A clever inventor robot with lightning motifs. Curious and talkative.", "english-default", _json.dumps(["drive", "light"])),
+        ("jaguar", "JAGUAR", "A sleek, powerful predator robot. Confident and agile.", "english-default", _json.dumps(["drive", "pounce"])),
+        ("vixen", "VIXEN (BMW)", "A cunning, stylish urban robot. Witty and resourceful.", "english-default", _json.dumps(["drive", "navigate"])),
+        ("panda", "PANDA", "A friendly, cuddly panda robot. Gentle and helpful.", "english-default", _json.dumps(["drive", "wave"])),
+    ]
+    for preset in presets:
+        await db.execute(
+            """INSERT OR IGNORE INTO character_presets
+               (id, name, description, system_prompt, voice_stack_id, motors)
+               VALUES (?, ?, ?, ?, ?, ?)""",
+            (preset[0], preset[1], preset[2], None, preset[3], preset[4]),
+        )
+    await db.commit()
+
+
 async def init_db():
     """Initialize SQLite database"""
     os.makedirs(os.path.dirname(DB_PATH), exist_ok=True)
@@ -283,6 +333,53 @@ async def init_db():
                 actor TEXT,
                 details TEXT
             );
+
+            -- Scheduler-managed voice stacks (STT/LLM/TTS configuration)
+            CREATE TABLE IF NOT EXISTS voice_stacks (
+                id TEXT PRIMARY KEY,
+                name TEXT NOT NULL,
+                stt_provider TEXT DEFAULT 'speaches',
+                stt_model TEXT DEFAULT 'Systran/faster-whisper-small',
+                stt_language TEXT DEFAULT 'en',
+                llm_provider TEXT DEFAULT 'ollama',
+                llm_model TEXT DEFAULT 'gemma4:e4b',
+                tts_provider TEXT DEFAULT 'elevenlabs',
+                tts_voice TEXT DEFAULT '21m00Tcm4TlvDq8ikWAM',
+                tts_language TEXT DEFAULT 'en',
+                allow_interruptions INTEGER DEFAULT 1,
+                min_endpointing_delay REAL DEFAULT 0.7,
+                max_turns INTEGER DEFAULT 20,
+                wake_word_enabled INTEGER DEFAULT 0,
+                created_at TEXT DEFAULT CURRENT_TIMESTAMP,
+                updated_at TEXT DEFAULT CURRENT_TIMESTAMP
+            );
+
+            -- Scheduler-managed character presets (production park characters)
+            CREATE TABLE IF NOT EXISTS character_presets (
+                id TEXT PRIMARY KEY,
+                name TEXT NOT NULL,
+                description TEXT,
+                system_prompt TEXT,
+                voice_stack_id TEXT,
+                motors TEXT,  -- JSON array
+                created_at TEXT DEFAULT CURRENT_TIMESTAMP,
+                updated_at TEXT DEFAULT CURRENT_TIMESTAMP,
+                FOREIGN KEY (voice_stack_id) REFERENCES voice_stacks(id)
+            );
+
+            -- Per-robot/device effective voice config override.
+            -- Falls back to the assigned character preset's voice stack.
+            CREATE TABLE IF NOT EXISTS robot_voice_configs (
+                robot_id TEXT PRIMARY KEY,
+                character_preset_id TEXT,
+                voice_stack_id TEXT,
+                override_config TEXT,  -- JSON merge on top of voice_stack
+                updated_at TEXT DEFAULT CURRENT_TIMESTAMP,
+                FOREIGN KEY (robot_id) REFERENCES robots(id) ON DELETE CASCADE,
+                FOREIGN KEY (character_preset_id) REFERENCES character_presets(id),
+                FOREIGN KEY (voice_stack_id) REFERENCES voice_stacks(id)
+            );
+
             CREATE INDEX IF NOT EXISTS idx_history_category ON history_log(category);
             CREATE INDEX IF NOT EXISTS idx_history_entity ON history_log(entity_type, entity_id);
         """)
@@ -317,6 +414,9 @@ async def init_db():
         # fleet view honest — only real connected hardware appears.
         await db.commit()
 
+        # Seed default voice stacks and production park character presets.
+        await _seed_voice_stacks_and_characters(db)
+
         # Seed default enrollment token + production_mode flag (only on first run)
         async with db.execute("SELECT value FROM settings WHERE key = 'default_enrollment_token'") as c:
             row = await c.fetchone()
@@ -2092,12 +2192,19 @@ async def device_request_session(device_id: str,
         .to_jwt()
     )
 
+    # Resolve the per-robot voice/character config so the ROBOVOICE agent and the
+    # robot client can apply it for this session. The agent fetches it via
+    # /api/devices/by-room/{room_name}/voice-config; we include it here too so
+    # the robot client can log/adjust local settings if needed.
+    voice_config = await _resolve_robot_voice_config(device_id)
+
     return {
         "session_id": session_id,
         "server_id": server["id"],
         "server_url": server["url"],
         "room_name": room_name,
         "token": token,
+        "voice_config": voice_config.model_dump() if voice_config else None,
     }
 
 # =============================================================================
@@ -2117,6 +2224,67 @@ class Character(CharacterSummary):
     system_prompt: Optional[str] = None
     raw: dict = {}
 
+# Scheduler-managed character presets (production park characters + custom).
+class CharacterPreset(BaseModel):
+    id: str
+    name: str
+    description: Optional[str] = None
+    system_prompt: Optional[str] = None
+    voice_stack_id: Optional[str] = None
+    motors: List[str] = []
+    created_at: Optional[datetime] = None
+    updated_at: Optional[datetime] = None
+
+class CharacterPresetCreate(BaseModel):
+    id: Optional[str] = None
+    name: str
+    description: Optional[str] = None
+    system_prompt: Optional[str] = None
+    voice_stack_id: Optional[str] = None
+    motors: List[str] = []
+
+class VoiceStack(BaseModel):
+    id: str
+    name: str
+    stt_provider: str = "speaches"
+    stt_model: str = "Systran/faster-whisper-small"
+    stt_language: str = "en"
+    llm_provider: str = "ollama"
+    llm_model: str = "gemma4:e4b"
+    tts_provider: str = "elevenlabs"
+    tts_voice: str = "21m00Tcm4TlvDq8ikWAM"
+    tts_language: str = "en"
+    allow_interruptions: bool = True
+    min_endpointing_delay: float = 0.7
+    max_turns: int = 20
+    wake_word_enabled: bool = False
+    created_at: Optional[datetime] = None
+    updated_at: Optional[datetime] = None
+
+class VoiceStackCreate(BaseModel):
+    id: Optional[str] = None
+    name: str
+    stt_provider: str = "speaches"
+    stt_model: str = "Systran/faster-whisper-small"
+    stt_language: str = "en"
+    llm_provider: str = "ollama"
+    llm_model: str = "gemma4:e4b"
+    tts_provider: str = "elevenlabs"
+    tts_voice: str = "21m00Tcm4TlvDq8ikWAM"
+    tts_language: str = "en"
+    allow_interruptions: bool = True
+    min_endpointing_delay: float = 0.7
+    max_turns: int = 20
+    wake_word_enabled: bool = False
+
+# Effective configuration returned to the ROBOVOICE agent for a given room/robot.
+class RobotVoiceConfig(BaseModel):
+    character_id: Optional[str] = None
+    character_name: Optional[str] = None
+    system_prompt: Optional[str] = None
+    voice_stack_id: Optional[str] = None
+    voice_stack: Optional[VoiceStack] = None
+
 def _load_robovoice_settings() -> dict:
     """Read the ROBOVOICE settings file. Returns {} if missing/invalid."""
     try:
@@ -2169,6 +2337,348 @@ async def get_character(character_id: str):
             )
     raise HTTPException(404, f"Character '{character_id}' not found in {ROBOVOICE_SETTINGS_PATH}")
 
+
+# ── Scheduler-managed character presets + voice stacks ─────────────────────
+
+async def _resolve_voice_stack_for_robot(db, robot_id: str) -> dict:
+    """Return the effective voice-stack dict for a robot/device id.
+
+    Resolution order:
+      1. robot_voice_configs.voice_stack_id override
+      2. character_presets.voice_stack_id assigned to the robot
+      3. 'english-default' fallback
+    """
+    # 1. robot_voice_configs row
+    async with db.execute(
+        "SELECT character_preset_id, voice_stack_id, override_config FROM robot_voice_configs WHERE robot_id = ?",
+        (robot_id,),
+    ) as c:
+        rvc = await c.fetchone()
+
+    preset_id = None
+    voice_stack_id = None
+    override = {}
+    if rvc:
+        preset_id = rvc["character_preset_id"]
+        voice_stack_id = rvc["voice_stack_id"]
+        if rvc["override_config"]:
+            try:
+                override = _json.loads(rvc["override_config"])
+            except Exception:
+                pass
+
+    # 2. If no explicit voice_stack, fall back via character preset
+    if not voice_stack_id:
+        if not preset_id:
+            async with db.execute(
+                "SELECT character_id FROM robots WHERE id = ? UNION ALL SELECT character_id FROM devices WHERE id = ?",
+                (robot_id, robot_id),
+            ) as c:
+                row = await c.fetchone()
+                if row:
+                    preset_id = row["character_id"]
+        if preset_id:
+            async with db.execute(
+                "SELECT voice_stack_id FROM character_presets WHERE id = ?", (preset_id,)
+            ) as c:
+                row = await c.fetchone()
+                if row:
+                    voice_stack_id = row["voice_stack_id"]
+
+    if not voice_stack_id:
+        voice_stack_id = "english-default"
+
+    async with db.execute(
+        "SELECT * FROM voice_stacks WHERE id = ?", (voice_stack_id,)
+    ) as c:
+        stack = await c.fetchone()
+    if not stack:
+        return {}
+
+    result = dict(stack)
+    result.update(override)
+    # SQLite stores booleans as 0/1; normalize
+    for key in ("allow_interruptions", "wake_word_enabled"):
+        if key in result:
+            result[key] = bool(result[key])
+    return result
+
+
+async def _resolve_robot_voice_config(robot_id: str) -> RobotVoiceConfig:
+    """Public helper to resolve the effective voice config for a robot id."""
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        stack_dict = await _resolve_voice_stack_for_robot(db, robot_id)
+
+        # Determine character preset + name
+        character_id = None
+        character_name = None
+        system_prompt = None
+        async with db.execute(
+            "SELECT character_preset_id FROM robot_voice_configs WHERE robot_id = ?", (robot_id,)
+        ) as c:
+            rvc = await c.fetchone()
+        if rvc and rvc["character_preset_id"]:
+            character_id = rvc["character_preset_id"]
+        else:
+            async with db.execute(
+                "SELECT character_id FROM robots WHERE id = ? UNION ALL SELECT character_id FROM devices WHERE id = ?",
+                (robot_id, robot_id),
+            ) as c:
+                row = await c.fetchone()
+                if row:
+                    character_id = row["character_id"]
+        if character_id:
+            async with db.execute(
+                "SELECT name, description, system_prompt FROM character_presets WHERE id = ?", (character_id,)
+            ) as c:
+                row = await c.fetchone()
+                if row:
+                    character_name = row["name"]
+                    system_prompt = row["system_prompt"]
+            if not character_name:
+                async with db.execute(
+                    "SELECT name FROM robots WHERE id = ? UNION ALL SELECT name FROM devices WHERE id = ?",
+                    (robot_id, robot_id),
+                ) as c:
+                    row = await c.fetchone()
+                    if row:
+                        character_name = row["name"]
+
+        voice_stack = VoiceStack(**stack_dict) if stack_dict else None
+        return RobotVoiceConfig(
+            character_id=character_id,
+            character_name=character_name,
+            system_prompt=system_prompt,
+            voice_stack_id=voice_stack.id if voice_stack else None,
+            voice_stack=voice_stack,
+        )
+
+
+@app.get("/api/voice-stacks", response_model=List[VoiceStack])
+async def list_voice_stacks():
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        async with db.execute("SELECT * FROM voice_stacks ORDER BY name") as c:
+            rows = await c.fetchall()
+            return [VoiceStack(**dict(r)) for r in rows]
+
+
+@app.post("/api/voice-stacks", response_model=VoiceStack)
+async def create_voice_stack(payload: VoiceStackCreate):
+    stack_id = payload.id or secrets.token_urlsafe(12)
+    now = datetime.utcnow().isoformat()
+    async with aiosqlite.connect(DB_PATH) as db:
+        await db.execute(
+            """INSERT INTO voice_stacks
+               (id, name, stt_provider, stt_model, stt_language, llm_provider, llm_model,
+                tts_provider, tts_voice, tts_language, allow_interruptions,
+                min_endpointing_delay, max_turns, wake_word_enabled, created_at, updated_at)
+               VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)""",
+            (
+                stack_id, payload.name, payload.stt_provider, payload.stt_model, payload.stt_language,
+                payload.llm_provider, payload.llm_model, payload.tts_provider, payload.tts_voice,
+                payload.tts_language, int(payload.allow_interruptions), payload.min_endpointing_delay,
+                payload.max_turns, int(payload.wake_word_enabled), now, now,
+            ),
+        )
+        await db.commit()
+    return await _get_voice_stack(stack_id)
+
+
+async def _get_voice_stack(stack_id: str) -> VoiceStack:
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        async with db.execute("SELECT * FROM voice_stacks WHERE id = ?", (stack_id,)) as c:
+            row = await c.fetchone()
+            if not row:
+                raise HTTPException(404, "Voice stack not found")
+            return VoiceStack(**dict(row))
+
+
+@app.get("/api/voice-stacks/{stack_id}", response_model=VoiceStack)
+async def get_voice_stack(stack_id: str):
+    return await _get_voice_stack(stack_id)
+
+
+@app.put("/api/voice-stacks/{stack_id}", response_model=VoiceStack)
+async def update_voice_stack(stack_id: str, payload: VoiceStackCreate):
+    now = datetime.utcnow().isoformat()
+    async with aiosqlite.connect(DB_PATH) as db:
+        async with db.execute("SELECT id FROM voice_stacks WHERE id = ?", (stack_id,)) as c:
+            if not await c.fetchone():
+                raise HTTPException(404, "Voice stack not found")
+        await db.execute(
+            """UPDATE voice_stacks SET
+               name = ?, stt_provider = ?, stt_model = ?, stt_language = ?, llm_provider = ?,
+               llm_model = ?, tts_provider = ?, tts_voice = ?, tts_language = ?,
+               allow_interruptions = ?, min_endpointing_delay = ?, max_turns = ?,
+               wake_word_enabled = ?, updated_at = ?
+               WHERE id = ?""",
+            (
+                payload.name, payload.stt_provider, payload.stt_model, payload.stt_language,
+                payload.llm_provider, payload.llm_model, payload.tts_provider, payload.tts_voice,
+                payload.tts_language, int(payload.allow_interruptions), payload.min_endpointing_delay,
+                payload.max_turns, int(payload.wake_word_enabled), now, stack_id,
+            ),
+        )
+        await db.commit()
+    return await _get_voice_stack(stack_id)
+
+
+@app.delete("/api/voice-stacks/{stack_id}")
+async def delete_voice_stack(stack_id: str):
+    async with aiosqlite.connect(DB_PATH) as db:
+        await db.execute("DELETE FROM voice_stacks WHERE id = ?", (stack_id,))
+        await db.commit()
+    return {"ok": True}
+
+
+@app.get("/api/character-presets", response_model=List[CharacterPreset])
+async def list_character_presets():
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        async with db.execute("SELECT * FROM character_presets ORDER BY name") as c:
+            rows = await c.fetchall()
+            return [
+                CharacterPreset(
+                    id=r["id"],
+                    name=r["name"],
+                    description=r["description"],
+                    system_prompt=r["system_prompt"],
+                    voice_stack_id=r["voice_stack_id"],
+                    motors=_json.loads(r["motors"]) if r["motors"] else [],
+                    created_at=r["created_at"],
+                    updated_at=r["updated_at"],
+                )
+                for r in rows
+            ]
+
+
+@app.post("/api/character-presets", response_model=CharacterPreset)
+async def create_character_preset(payload: CharacterPresetCreate):
+    preset_id = payload.id or secrets.token_urlsafe(12)
+    now = datetime.utcnow().isoformat()
+    async with aiosqlite.connect(DB_PATH) as db:
+        await db.execute(
+            """INSERT INTO character_presets
+               (id, name, description, system_prompt, voice_stack_id, motors, created_at, updated_at)
+               VALUES (?, ?, ?, ?, ?, ?, ?, ?)""",
+            (
+                preset_id, payload.name, payload.description, payload.system_prompt,
+                payload.voice_stack_id, _json.dumps(payload.motors or []), now, now,
+            ),
+        )
+        await db.commit()
+    return await _get_character_preset(preset_id)
+
+
+async def _get_character_preset(preset_id: str) -> CharacterPreset:
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        async with db.execute("SELECT * FROM character_presets WHERE id = ?", (preset_id,)) as c:
+            row = await c.fetchone()
+            if not row:
+                raise HTTPException(404, "Character preset not found")
+            return CharacterPreset(
+                id=row["id"],
+                name=row["name"],
+                description=row["description"],
+                system_prompt=row["system_prompt"],
+                voice_stack_id=row["voice_stack_id"],
+                motors=_json.loads(row["motors"]) if row["motors"] else [],
+                created_at=row["created_at"],
+                updated_at=row["updated_at"],
+            )
+
+
+@app.get("/api/character-presets/{preset_id}", response_model=CharacterPreset)
+async def get_character_preset(preset_id: str):
+    return await _get_character_preset(preset_id)
+
+
+@app.put("/api/character-presets/{preset_id}", response_model=CharacterPreset)
+async def update_character_preset(preset_id: str, payload: CharacterPresetCreate):
+    now = datetime.utcnow().isoformat()
+    async with aiosqlite.connect(DB_PATH) as db:
+        async with db.execute("SELECT id FROM character_presets WHERE id = ?", (preset_id,)) as c:
+            if not await c.fetchone():
+                raise HTTPException(404, "Character preset not found")
+        await db.execute(
+            """UPDATE character_presets SET
+               name = ?, description = ?, system_prompt = ?, voice_stack_id = ?, motors = ?, updated_at = ?
+               WHERE id = ?""",
+            (
+                payload.name, payload.description, payload.system_prompt,
+                payload.voice_stack_id, _json.dumps(payload.motors or []), now, preset_id,
+            ),
+        )
+        await db.commit()
+    return await _get_character_preset(preset_id)
+
+
+@app.delete("/api/character-presets/{preset_id}")
+async def delete_character_preset(preset_id: str):
+    async with aiosqlite.connect(DB_PATH) as db:
+        await db.execute("DELETE FROM character_presets WHERE id = ?", (preset_id,))
+        await db.commit()
+    return {"ok": True}
+
+
+class RobotVoiceConfigUpdate(BaseModel):
+    character_preset_id: Optional[str] = None
+    voice_stack_id: Optional[str] = None
+    override_config: Optional[dict] = None
+
+
+@app.get("/api/robots/{robot_id}/voice-config", response_model=RobotVoiceConfig)
+async def get_robot_voice_config(robot_id: str):
+    return await _resolve_robot_voice_config(robot_id)
+
+
+@app.put("/api/robots/{robot_id}/voice-config", response_model=RobotVoiceConfig)
+async def update_robot_voice_config(robot_id: str, payload: RobotVoiceConfigUpdate):
+    now = datetime.utcnow().isoformat()
+    async with aiosqlite.connect(DB_PATH) as db:
+        await db.execute(
+            """INSERT INTO robot_voice_configs (robot_id, character_preset_id, voice_stack_id, override_config, updated_at)
+               VALUES (?, ?, ?, ?, ?)
+               ON CONFLICT(robot_id) DO UPDATE SET
+                   character_preset_id = excluded.character_preset_id,
+                   voice_stack_id = excluded.voice_stack_id,
+                   override_config = excluded.override_config,
+                   updated_at = excluded.updated_at""",
+            (
+                robot_id,
+                payload.character_preset_id,
+                payload.voice_stack_id,
+                _json.dumps(payload.override_config) if payload.override_config else None,
+                now,
+            ),
+        )
+        await db.commit()
+    return await _resolve_robot_voice_config(robot_id)
+
+
+@app.get("/api/devices/by-room/{room_name}/voice-config", response_model=RobotVoiceConfig)
+async def get_device_voice_config_by_room(room_name: str):
+    """ROBOVOICE agent calls this when it joins a room to fetch per-robot config."""
+    async with aiosqlite.connect(DB_PATH) as db:
+        db.row_factory = aiosqlite.Row
+        async with db.execute(
+            "SELECT robot_id FROM sessions WHERE room_name = ? ORDER BY started_at DESC LIMIT 1",
+            (room_name,),
+        ) as c:
+            session = await c.fetchone()
+        if not session:
+            if room_name.startswith("robopark-"):
+                device_id = room_name[len("robopark-"):]
+                if device_id:
+                    return await _resolve_robot_voice_config(device_id)
+            raise HTTPException(404, "No session bound to this room")
+        return await _resolve_robot_voice_config(session["robot_id"])
+
+
 # =============================================================================
 # TELEMETRY (close-the-loop: trigger counts, latency, drop rate, error reasons)
 # =============================================================================
@@ -2357,11 +2867,24 @@ async def dashboard():
                     const deviceInventory = ref({});
                     const showLiveKitConfig = ref(false);
                     const livekitForm = ref({ url: '', api_key: '', api_secret: '' });
+                    const voiceStacks = ref([]);
+                    const characterPresets = ref([]);
+                    const showVoiceStackForm = ref(false);
+                    const showCharacterForm = ref(false);
+                    const editingVoiceStack = ref(null);
+                    const editingCharacter = ref(null);
+                    const voiceStackForm = ref({
+                        id: '', name: '', stt_provider: 'speaches', stt_model: 'Systran/faster-whisper-small', stt_language: 'en',
+                        llm_provider: 'ollama', llm_model: 'gemma4:e4b', tts_provider: 'elevenlabs',
+                        tts_voice: '21m00Tcm4TlvDq8ikWAM', tts_language: 'en',
+                        allow_interruptions: true, min_endpointing_delay: 0.7, max_turns: 20, wake_word_enabled: false,
+                    });
+                    const characterForm = ref({ id: '', name: '', description: '', system_prompt: '', voice_stack_id: '', motors: [], motorsStr: '' });
                     let ws = null;
 
                     const fetchData = async () => {
                         try {
-                            const [robotsRes, serversRes, sessionsRes, statsRes, devsRes, settingsRes, lkRes] = await Promise.all([
+                            const [robotsRes, serversRes, sessionsRes, statsRes, devsRes, settingsRes, lkRes, stacksRes, charsRes] = await Promise.all([
                                 fetch('/api/robots').then(r => r.json()),
                                 fetch('/api/servers').then(r => r.json()),
                                 fetch('/api/sessions?active_only=true').then(r => r.json()),
@@ -2369,6 +2892,8 @@ async def dashboard():
                                 fetch('/api/devices').then(r => r.json()),
                                 fetch('/api/settings').then(r => r.json()),
                                 fetch('/api/livekit/config').then(r => r.json()),
+                                fetch('/api/voice-stacks').then(r => r.json()).catch(() => []),
+                                fetch('/api/character-presets').then(r => r.json()).catch(() => []),
                             ]);
                             robots.value = robotsRes;
                             servers.value = serversRes;
@@ -2377,6 +2902,8 @@ async def dashboard():
                             devices.value = devsRes;
                             settings.value = settingsRes;
                             livekit.value = lkRes;
+                            voiceStacks.value = stacksRes;
+                            characterPresets.value = charsRes;
 
                             for (const server of serversRes) {
                                 try {
@@ -2689,15 +3216,112 @@ async def dashboard():
                         if (ws) ws.close();
                     });
 
+                    const openVoiceStackForm = (stack = null) => {
+                        if (stack) {
+                            editingVoiceStack.value = stack.id;
+                            voiceStackForm.value = { ...stack };
+                        } else {
+                            editingVoiceStack.value = null;
+                            voiceStackForm.value = {
+                                id: '', name: '', stt_provider: 'speaches', stt_model: 'Systran/faster-whisper-small', stt_language: 'en',
+                                llm_provider: 'ollama', llm_model: 'gemma4:e4b', tts_provider: 'elevenlabs',
+                                tts_voice: '21m00Tcm4TlvDq8ikWAM', tts_language: 'en',
+                                allow_interruptions: true, min_endpointing_delay: 0.7, max_turns: 20, wake_word_enabled: false,
+                            };
+                        }
+                        showVoiceStackForm.value = true;
+                    };
+
+                    const saveVoiceStack = async () => {
+                        try {
+                            const method = editingVoiceStack.value ? 'PUT' : 'POST';
+                            const url = editingVoiceStack.value ? `/api/voice-stacks/${editingVoiceStack.value}` : '/api/voice-stacks';
+                            await fetch(url, { method, headers: {'Content-Type': 'application/json'}, body: JSON.stringify(voiceStackForm.value) });
+                            showVoiceStackForm.value = false;
+                            fetchData();
+                        } catch (e) {
+                            alert('Failed to save voice stack: ' + e.message);
+                        }
+                    };
+
+                    const deleteVoiceStack = async (id) => {
+                        if (!confirm('Delete this voice stack? Any presets using it will need reassignment.')) return;
+                        try {
+                            await fetch(`/api/voice-stacks/${id}`, { method: 'DELETE' });
+                            fetchData();
+                        } catch (e) {
+                            alert('Failed to delete voice stack: ' + e.message);
+                        }
+                    };
+
+                    const openCharacterForm = (char = null) => {
+                        if (char) {
+                            editingCharacter.value = char.id;
+                            const motorsArr = Array.isArray(char.motors) ? char.motors : [];
+                            characterForm.value = { ...char, motors: motorsArr, motorsStr: motorsArr.join(', ') };
+                        } else {
+                            editingCharacter.value = null;
+                            characterForm.value = { id: '', name: '', description: '', system_prompt: '', voice_stack_id: '', motors: [], motorsStr: '' };
+                        }
+                        showCharacterForm.value = true;
+                    };
+
+                    const saveCharacter = async () => {
+                        try {
+                            const motors = characterForm.value.motorsStr
+                                ? characterForm.value.motorsStr.split(',').map(x => x.trim()).filter(Boolean)
+                                : (characterForm.value.motors || []);
+                            const payload = { ...characterForm.value, motors };
+                            delete payload.motorsStr;
+                            const method = editingCharacter.value ? 'PUT' : 'POST';
+                            const url = editingCharacter.value ? `/api/character-presets/${editingCharacter.value}` : '/api/character-presets';
+                            await fetch(url, { method, headers: {'Content-Type': 'application/json'}, body: JSON.stringify(payload) });
+                            showCharacterForm.value = false;
+                            fetchData();
+                        } catch (e) {
+                            alert('Failed to save character preset: ' + e.message);
+                        }
+                    };
+
+                    const deleteCharacter = async (id) => {
+                        if (!confirm('Delete this character preset?')) return;
+                        try {
+                            await fetch(`/api/character-presets/${id}`, { method: 'DELETE' });
+                            fetchData();
+                        } catch (e) {
+                            alert('Failed to delete character preset: ' + e.message);
+                        }
+                    };
+
+                    const assignRobotCharacter = async (robotId, characterId) => {
+                        try {
+                            await fetch(`/api/robots/${robotId}/voice-config`, {
+                                method: 'PUT', headers: {'Content-Type': 'application/json'},
+                                body: JSON.stringify({ character_id: characterId || null }),
+                            });
+                            fetchData();
+                        } catch (e) {
+                            alert('Failed to assign character: ' + e.message);
+                        }
+                    };
+
+                    const stackName = (id) => voiceStacks.value.find(s => s.id === id)?.name || '—';
+                    const characterName = (id) => characterPresets.value.find(c => c.id === id)?.name || '—';
+
                     return {
                         robots, servers, sessions, stats, connected, serverMetrics,
                         devices, settings, livekit, showAddDevice, showDevices, newDevice, rotatedToken,
                         showLiveKitConfig, livekitForm, preview, deviceInventory,
+                        voiceStacks, characterPresets, showVoiceStackForm, showCharacterForm,
+                        editingVoiceStack, editingCharacter, voiceStackForm, characterForm,
                         loadModel, unloadModel, statusColor, formatDuration, formatTime,
                         toggleProductionMode, rotateEnrollmentToken,
                         submitNewDevice, deleteDevice, rotateDeviceToken, dismissRotatedToken,
                         joinConversation, saveLiveKitConfig, openLiveKitConfig,
                         startPreview, stopPreview, updateDeviceMedia,
+                        openVoiceStackForm, saveVoiceStack, deleteVoiceStack,
+                        openCharacterForm, saveCharacter, deleteCharacter,
+                        assignRobotCharacter, stackName, characterName,
                     };
                 },
                 template: `
@@ -2948,6 +3572,25 @@ async def dashboard():
                                                 {{ robot.total_sessions }} sessions
                                             </div>
                                         </div>
+                                        <div class="text-xs text-gray-400 mb-2">
+                                            Character: <strong>{{ characterName(robot.character_id) }}</strong> ·
+                                            Voice Stack: <strong>{{ stackName(robot.voice_stack_id) }}</strong>
+                                        </div>
+                                        <div class="grid grid-cols-2 gap-2 mb-2 text-sm">
+                                            <div>
+                                                <label class="text-xs text-gray-500">Character</label>
+                                                <select :value="robot.character_id"
+                                                        @change="assignRobotCharacter(robot.id, $event.target.value)"
+                                                        class="w-full bg-gray-800 rounded px-2 py-1 text-xs">
+                                                    <option value="">—</option>
+                                                    <option v-for="c in characterPresets" :key="c.id" :value="c.id">{{ c.name }}</option>
+                                                </select>
+                                            </div>
+                                            <div>
+                                                <label class="text-xs text-gray-500">Voice Stack</label>
+                                                <div class="text-xs text-gray-400 px-2 py-1 bg-gray-800 rounded truncate">{{ stackName(robot.voice_stack_id) }}</div>
+                                            </div>
+                                        </div>
                                         <div v-if="deviceInventory[robot.id]" class="space-y-2 text-sm">
                                             <div class="grid grid-cols-2 gap-2">
                                                 <div>
@@ -2987,6 +3630,63 @@ async def dashboard():
                                     <div class="text-xs text-gray-400 px-2 py-1">{{ preview.room }}</div>
                                 </div>
                             </div>
+
+                            <!-- Voice Configuration Panel -->
+                            <div class="col-span-4 bg-gray-800 rounded-lg p-4">
+                                <div class="flex justify-between items-center mb-4">
+                                    <h2 class="text-xl font-semibold">🎙️ Voice Stacks</h2>
+                                    <button @click="openVoiceStackForm()" class="px-2 py-1 bg-cyan-700 hover:bg-cyan-600 rounded text-xs">+ New Stack</button>
+                                </div>
+                                <div class="space-y-3">
+                                    <div v-for="s in voiceStacks" :key="s.id" class="bg-gray-700 rounded-lg p-3">
+                                        <div class="flex justify-between items-start">
+                                            <div>
+                                                <div class="font-medium">{{ s.name }}</div>
+                                                <div class="text-xs text-gray-400 mt-1">
+                                                    STT <strong>{{ s.stt_provider }}/{{ s.stt_model }}</strong> · {{ s.stt_language }} ·
+                                                    LLM <strong>{{ s.llm_provider }}/{{ s.llm_model }}</strong> ·
+                                                    TTS <strong>{{ s.tts_provider }}</strong> · {{ s.tts_language }}
+                                                </div>
+                                                <div class="text-xs text-gray-500 mt-1">
+                                                    interruptions={{ s.allow_interruptions ? 'on' : 'off' }} · endpointing={{ s.min_endpointing_delay }}s · max_turns={{ s.max_turns }}
+                                                </div>
+                                            </div>
+                                            <div class="flex gap-1">
+                                                <button @click="openVoiceStackForm(s)" class="text-xs px-2 py-1 bg-gray-600 hover:bg-gray-500 rounded">Edit</button>
+                                                <button @click="deleteVoiceStack(s.id)" class="text-xs px-2 py-1 bg-red-700 hover:bg-red-600 rounded">×</button>
+                                            </div>
+                                        </div>
+                                    </div>
+                                    <div v-if="voiceStacks.length === 0" class="text-gray-500 text-center py-4 text-sm">
+                                        No voice stacks configured.
+                                    </div>
+                                </div>
+
+                                <div class="flex justify-between items-center mb-4 mt-6">
+                                    <h2 class="text-xl font-semibold">🎭 Character Presets</h2>
+                                    <button @click="openCharacterForm()" class="px-2 py-1 bg-cyan-700 hover:bg-cyan-600 rounded text-xs">+ New Character</button>
+                                </div>
+                                <div class="space-y-3">
+                                    <div v-for="c in characterPresets" :key="c.id" class="bg-gray-700 rounded-lg p-3">
+                                        <div class="flex justify-between items-start">
+                                            <div>
+                                                <div class="font-medium">{{ c.name }}</div>
+                                                <div class="text-xs text-gray-400 mt-1">{{ c.description }}</div>
+                                                <div class="text-xs text-gray-500 mt-1">
+                                                    stack: <strong>{{ stackName(c.voice_stack_id) }}</strong> · motors: {{ (c.motors || '').replace(/[^a-zA-Z,]/g, '') || '—' }}
+                                                </div>
+                                            </div>
+                                            <div class="flex gap-1">
+                                                <button @click="openCharacterForm(c)" class="text-xs px-2 py-1 bg-gray-600 hover:bg-gray-500 rounded">Edit</button>
+                                                <button @click="deleteCharacter(c.id)" class="text-xs px-2 py-1 bg-red-700 hover:bg-red-600 rounded">×</button>
+                                            </div>
+                                        </div>
+                                    </div>
+                                    <div v-if="characterPresets.length === 0" class="text-gray-500 text-center py-4 text-sm">
+                                        No character presets configured.
+                                    </div>
+                                </div>
+                            </div>
                             
                             <!-- Servers Panel -->
                             <div class="col-span-5 space-y-4">
@@ -3054,10 +3754,135 @@ async def dashboard():
                                         <div class="text-gray-400 text-xs">
                                             {{ session.server_id }} • {{ session.room_name }}
                                         </div>
+                                        <div v-if="session.voice_config" class="text-xs text-cyan-400 mt-1">
+                                            {{ session.voice_config.character_id }} · {{ session.voice_config.voice_stack_id }}
+                                        </div>
                                     </div>
                                 </div>
                             </div>
                         </div>
+
+                        <!-- Voice Stack Modal -->
+                        <div v-if="showVoiceStackForm" class="fixed inset-0 bg-black/70 flex items-center justify-center z-50" @click.self="showVoiceStackForm = false">
+                            <div class="bg-gray-800 rounded-lg p-6 w-full max-w-2xl max-h-[90vh] overflow-y-auto">
+                                <h3 class="text-lg font-semibold mb-2">{{ editingVoiceStack ? 'Edit Voice Stack' : 'New Voice Stack' }}</h3>
+                                <div class="grid grid-cols-2 gap-3 text-sm">
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">ID</label>
+                                        <input v-model="voiceStackForm.id" :disabled="!!editingVoiceStack" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="e.g. english-fast">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">Name</label>
+                                        <input v-model="voiceStackForm.name" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="English Fast">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">STT Provider</label>
+                                        <select v-model="voiceStackForm.stt_provider" class="w-full bg-gray-700 rounded px-3 py-2">
+                                            <option>speaches</option>
+                                            <option>deepgram</option>
+                                            <option>openai</option>
+                                        </select>
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">STT Model</label>
+                                        <input v-model="voiceStackForm.stt_model" class="w-full bg-gray-700 rounded px-3 py-2">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">STT Language</label>
+                                        <input v-model="voiceStackForm.stt_language" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="en/he">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">LLM Provider</label>
+                                        <select v-model="voiceStackForm.llm_provider" class="w-full bg-gray-700 rounded px-3 py-2">
+                                            <option>ollama</option>
+                                            <option>openai</option>
+                                        </select>
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">LLM Model</label>
+                                        <input v-model="voiceStackForm.llm_model" class="w-full bg-gray-700 rounded px-3 py-2">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">TTS Provider</label>
+                                        <select v-model="voiceStackForm.tts_provider" class="w-full bg-gray-700 rounded px-3 py-2">
+                                            <option>elevenlabs</option>
+                                            <option>speaches</option>
+                                            <option>cartesia</option>
+                                        </select>
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">TTS Voice ID</label>
+                                        <input v-model="voiceStackForm.tts_voice" class="w-full bg-gray-700 rounded px-3 py-2">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">TTS Language</label>
+                                        <input v-model="voiceStackForm.tts_language" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="en">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">Min Endpointing Delay (s)</label>
+                                        <input v-model.number="voiceStackForm.min_endpointing_delay" type="number" step="0.1" class="w-full bg-gray-700 rounded px-3 py-2">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">Max Turns</label>
+                                        <input v-model.number="voiceStackForm.max_turns" type="number" class="w-full bg-gray-700 rounded px-3 py-2">
+                                    </div>
+                                    <div class="flex items-center gap-2">
+                                        <input id="vs_interruptions" v-model="voiceStackForm.allow_interruptions" type="checkbox" class="rounded">
+                                        <label for="vs_interruptions" class="text-sm">Allow Interruptions</label>
+                                    </div>
+                                    <div class="flex items-center gap-2">
+                                        <input id="vs_wake" v-model="voiceStackForm.wake_word_enabled" type="checkbox" class="rounded">
+                                        <label for="vs_wake" class="text-sm">Wake Word Enabled</label>
+                                    </div>
+                                </div>
+                                <div class="flex justify-end gap-2 mt-5">
+                                    <button @click="showVoiceStackForm = false" class="px-3 py-2 bg-gray-700 hover:bg-gray-600 rounded">Cancel</button>
+                                    <button @click="saveVoiceStack" class="px-3 py-2 bg-cyan-600 hover:bg-cyan-700 rounded">Save</button>
+                                </div>
+                            </div>
+                        </div>
+
+                        <!-- Character Modal -->
+                        <div v-if="showCharacterForm" class="fixed inset-0 bg-black/70 flex items-center justify-center z-50" @click.self="showCharacterForm = false">
+                            <div class="bg-gray-800 rounded-lg p-6 w-full max-w-2xl max-h-[90vh] overflow-y-auto">
+                                <h3 class="text-lg font-semibold mb-2">{{ editingCharacter ? 'Edit Character' : 'New Character' }}</h3>
+                                <div class="space-y-3 text-sm">
+                                    <div class="grid grid-cols-2 gap-3">
+                                        <div>
+                                            <label class="block text-gray-400 mb-1">ID</label>
+                                            <input v-model="characterForm.id" :disabled="!!editingCharacter" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="volt">
+                                        </div>
+                                        <div>
+                                            <label class="block text-gray-400 mb-1">Name</label>
+                                            <input v-model="characterForm.name" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="VOLT">
+                                        </div>
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">Voice Stack</label>
+                                        <select v-model="characterForm.voice_stack_id" class="w-full bg-gray-700 rounded px-3 py-2">
+                                            <option value="">—</option>
+                                            <option v-for="s in voiceStacks" :key="s.id" :value="s.id">{{ s.name }}</option>
+                                        </select>
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">Description</label>
+                                        <input v-model="characterForm.description" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="Short public description">
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">System Prompt</label>
+                                        <textarea v-model="characterForm.system_prompt" rows="8" class="w-full bg-gray-700 rounded px-3 py-2 font-mono" placeholder="You are a friendly robot..."></textarea>
+                                    </div>
+                                    <div>
+                                        <label class="block text-gray-400 mb-1">Motor Commands (comma-separated)</label>
+                                        <input v-model="characterForm.motorsStr" class="w-full bg-gray-700 rounded px-3 py-2" placeholder="drive, navigate, stop">
+                                    </div>
+                                </div>
+                                <div class="flex justify-end gap-2 mt-5">
+                                    <button @click="showCharacterForm = false" class="px-3 py-2 bg-gray-700 hover:bg-gray-600 rounded">Cancel</button>
+                                    <button @click="saveCharacter" class="px-3 py-2 bg-cyan-600 hover:bg-cyan-700 rounded">Save</button>
+                                </div>
+                            </div>
+                        </div>
                     </div>
                 `
             }).mount('#app');
@@ -3076,7 +3901,7 @@ async def robot_client_page():
     detection, session triggering, and dashboard streaming before deploying to
     real Pis.
     """
-    return """
+    return r"""
 <!DOCTYPE html>
 <html>
 <head>