Add Settings subclasses to all services and auto-discovered init tests

- Add dedicated Settings subclasses to 20 LLM services that were
  borrowing parent Settings classes (e.g. AzureLLMSettings,
  GroqLLMSettings) so users don't need cross-module imports
- Fix field defaults to NOT_GIVEN in BaseWhisperSTTSettings,
  OpenAIRealtimeSTTSettings, and NvidiaSegmentedSTTSettings for
  delta-mode safety
- Fix incomplete default_settings in AWS, Cartesia, ElevenLabs,
  Fish, and Whisper services so validate_complete() passes
- Add auto-discovered tests that verify all Settings classes default
  to NOT_GIVEN (delta safety) and all services initialize with
  complete settings (store completeness)

src/pipecat/services/aws/stt.py

@@ -98,6 +98,7 @@ class AWSTranscribeSTTService(WebsocketSTTService):
         """
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = AWSTranscribeSTTSettings(
+            model=None,
             language=self.language_to_service_language(Language.EN) or "en-US",
         )
 

src/pipecat/services/azure/llm.py

@@ -6,6 +6,7 @@
 
 """Azure OpenAI service implementation for the Pipecat AI framework."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -16,6 +17,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class AzureLLMSettings(OpenAILLMSettings):
+    """Settings for Azure OpenAI LLM service."""
+
+    pass
+
+
 class AzureLLMService(OpenAILLMService):
     """A service for interacting with Azure OpenAI using the OpenAI-compatible interface.
 
@@ -30,7 +38,7 @@ class AzureLLMService(OpenAILLMService):
         endpoint: str,
         model: Optional[str] = None,
         api_version: str = "2024-09-01-preview",
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[AzureLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the Azure LLM service.
@@ -49,11 +57,11 @@ class AzureLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="gpt-4o")
+        default_settings = AzureLLMSettings(model="gpt-4o")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", AzureLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/azure/realtime/llm.py

@@ -6,9 +6,11 @@
 
 """Azure OpenAI Realtime LLM service implementation."""
 
+from dataclasses import dataclass
+
 from loguru import logger
 
-from pipecat.services.openai.realtime.llm import OpenAIRealtimeLLMService
+from pipecat.services.openai.realtime.llm import OpenAIRealtimeLLMService, OpenAIRealtimeLLMSettings
 
 try:
     from websockets.asyncio.client import connect as websocket_connect
@@ -18,6 +20,13 @@ except ModuleNotFoundError as e:
     raise Exception(f"Missing module: {e}")
 
 
+@dataclass
+class AzureRealtimeLLMSettings(OpenAIRealtimeLLMSettings):
+    """Settings for Azure Realtime LLM service."""
+
+    pass
+
+
 class AzureRealtimeLLMService(OpenAIRealtimeLLMService):
     """Azure OpenAI Realtime LLM service with Azure-specific authentication.
 

src/pipecat/services/cartesia/tts.py

@@ -301,6 +301,7 @@ class CartesiaTTSService(AudioContextTTSService):
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = CartesiaTTSSettings(
             model="sonic-3",
+            voice=None,
             language=language_to_cartesia_language(Language.EN),
             generation_config=None,
             pronunciation_dict_id=None,
@@ -745,6 +746,7 @@ class CartesiaHttpTTSService(TTSService):
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = CartesiaTTSSettings(
             model="sonic-3",
+            voice=None,
             language=language_to_cartesia_language(Language.EN),
             generation_config=None,
             pronunciation_dict_id=None,

src/pipecat/services/cerebras/llm.py

@@ -6,6 +6,7 @@
 
 """Cerebras LLM service implementation using OpenAI-compatible interface."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -16,6 +17,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class CerebrasLLMSettings(OpenAILLMSettings):
+    """Settings for Cerebras LLM service."""
+
+    pass
+
+
 class CerebrasLLMService(OpenAILLMService):
     """A service for interacting with Cerebras's API using the OpenAI-compatible interface.
 
@@ -29,7 +37,7 @@ class CerebrasLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://api.cerebras.ai/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[CerebrasLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the Cerebras LLM service.
@@ -47,11 +55,11 @@ class CerebrasLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="gpt-oss-120b")
+        default_settings = CerebrasLLMSettings(model="gpt-oss-120b")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", CerebrasLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/deepseek/llm.py

@@ -6,6 +6,7 @@
 
 """DeepSeek LLM service implementation using OpenAI-compatible interface."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -16,6 +17,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class DeepSeekLLMSettings(OpenAILLMSettings):
+    """Settings for DeepSeek LLM service."""
+
+    pass
+
+
 class DeepSeekLLMService(OpenAILLMService):
     """A service for interacting with DeepSeek's API using the OpenAI-compatible interface.
 
@@ -29,7 +37,7 @@ class DeepSeekLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://api.deepseek.com/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[DeepSeekLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the DeepSeek LLM service.
@@ -47,11 +55,11 @@ class DeepSeekLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="deepseek-chat")
+        default_settings = DeepSeekLLMSettings(model="deepseek-chat")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", DeepSeekLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/elevenlabs/tts.py

@@ -418,6 +418,14 @@ class ElevenLabsTTSService(AudioContextTTSService):
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = ElevenLabsTTSSettings(
             model="eleven_turbo_v2_5",
+            voice=None,
+            language=None,
+            stability=None,
+            similarity_boost=None,
+            style=None,
+            use_speaker_boost=None,
+            speed=None,
+            apply_text_normalization=None,
         )
 
         # Track init-only URL params through the override chain
@@ -986,6 +994,15 @@ class ElevenLabsHttpTTSService(TTSService):
         # 1. Initialize default_settings with hardcoded defaults
         default_settings = ElevenLabsHttpTTSSettings(
             model="eleven_turbo_v2_5",
+            voice=None,
+            language=None,
+            optimize_streaming_latency=None,
+            stability=None,
+            similarity_boost=None,
+            style=None,
+            use_speaker_boost=None,
+            speed=None,
+            apply_text_normalization=None,
         )
 
         # 2. Apply direct init arg overrides (deprecated)

src/pipecat/services/fireworks/llm.py

@@ -6,6 +6,7 @@
 
 """Fireworks AI service implementation using OpenAI-compatible interface."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -16,6 +17,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class FireworksLLMSettings(OpenAILLMSettings):
+    """Settings for Fireworks LLM service."""
+
+    pass
+
+
 class FireworksLLMService(OpenAILLMService):
     """A service for interacting with Fireworks AI using the OpenAI-compatible interface.
 
@@ -29,7 +37,7 @@ class FireworksLLMService(OpenAILLMService):
         api_key: str,
         model: Optional[str] = None,
         base_url: str = "https://api.fireworks.ai/inference/v1",
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[FireworksLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the Fireworks LLM service.
@@ -47,11 +55,11 @@ class FireworksLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="accounts/fireworks/models/firefunction-v2")
+        default_settings = FireworksLLMSettings(model="accounts/fireworks/models/firefunction-v2")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", FireworksLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/fish/tts.py

@@ -173,6 +173,7 @@ class FishAudioTTSService(InterruptibleTTSService):
         default_settings = FishAudioTTSSettings(
             model="s1",
             voice=None,
+            language=None,
             latency="normal",
             normalize=True,
             prosody_speed=1.0,

src/pipecat/services/google/gemini_live/llm_vertex.py

@@ -12,6 +12,7 @@ streaming responses, and tool usage.
 """
 
 import json
+from dataclasses import dataclass
 from typing import List, Optional, Union
 
 from loguru import logger
@@ -41,6 +42,13 @@ except ModuleNotFoundError as e:
     raise Exception(f"Missing module: {e}")
 
 
+@dataclass
+class GeminiLiveVertexLLMSettings(GeminiLiveLLMSettings):
+    """Settings for Gemini Live Vertex LLM service."""
+
+    pass
+
+
 class GeminiLiveVertexLLMService(GeminiLiveLLMService):
     """Provides access to Google's Gemini Live model via Vertex AI.
 
@@ -63,7 +71,7 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
         system_instruction: Optional[str] = None,
         tools: Optional[Union[List[dict], ToolsSchema]] = None,
         params: Optional[InputParams] = None,
-        settings: Optional[GeminiLiveLLMSettings] = None,
+        settings: Optional[GeminiLiveVertexLLMSettings] = None,
         inference_on_context_initialization: bool = True,
         file_api_base_url: str = "https://generativelanguage.googleapis.com/v1beta/files",
         http_options: Optional[HttpOptions] = None,
@@ -122,7 +130,7 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
         # double deprecation warnings from the parent.
 
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = GeminiLiveLLMSettings(
+        default_settings = GeminiLiveVertexLLMSettings(
             model="google/gemini-live-2.5-flash-native-audio",
             frequency_penalty=None,
             max_tokens=4096,
@@ -146,12 +154,12 @@ class GeminiLiveVertexLLMService(GeminiLiveLLMService):
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", GeminiLiveLLMSettings, "model")
+            _warn_deprecated_param("model", GeminiLiveVertexLLMSettings, "model")
             default_settings.model = model
 
         # 3. Apply params overrides — only if settings not provided
         if params is not None:
-            _warn_deprecated_param("params", GeminiLiveLLMSettings)
+            _warn_deprecated_param("params", GeminiLiveVertexLLMSettings)
             if not settings:
                 default_settings.frequency_penalty = params.frequency_penalty
                 default_settings.max_tokens = params.max_tokens

src/pipecat/services/google/llm_openai.py

@@ -12,6 +12,7 @@ API format through Google's Gemini API OpenAI compatibility layer.
 
 import json
 import os
+from dataclasses import dataclass
 from typing import Optional
 
 from openai import AsyncStream
@@ -32,6 +33,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class GoogleOpenAILLMSettings(OpenAILLMSettings):
+    """Settings for Google OpenAI-compatible LLM service."""
+
+    pass
+
+
 class GoogleLLMOpenAIBetaService(OpenAILLMService):
     """Google LLM service using OpenAI-compatible API format.
 
@@ -56,7 +64,7 @@ class GoogleLLMOpenAIBetaService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://generativelanguage.googleapis.com/v1beta/openai/",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[GoogleOpenAILLMSettings] = None,
         **kwargs,
     ):
         """Initialize the Google LLM service.
@@ -85,11 +93,11 @@ class GoogleLLMOpenAIBetaService(OpenAILLMService):
             )
 
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="gemini-2.0-flash")
+        default_settings = GoogleOpenAILLMSettings(model="gemini-2.0-flash")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", GoogleOpenAILLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/google/llm_vertex.py

@@ -12,6 +12,7 @@ extending the GoogleLLMService with Vertex AI authentication.
 
 import json
 import os
+from dataclasses import dataclass
 
 # Suppress gRPC fork warnings
 os.environ["GRPC_ENABLE_FORK_SUPPORT"] = "false"
@@ -39,6 +40,13 @@ except ModuleNotFoundError as e:
     raise Exception(f"Missing module: {e}")
 
 
+@dataclass
+class GoogleVertexLLMSettings(GoogleLLMSettings):
+    """Settings for Google Vertex LLM service."""
+
+    pass
+
+
 class GoogleVertexLLMService(GoogleLLMService):
     """Google Vertex AI LLM service extending GoogleLLMService.
 
@@ -104,7 +112,7 @@ class GoogleVertexLLMService(GoogleLLMService):
         location: Optional[str] = None,
         project_id: Optional[str] = None,
         params: Optional[GoogleLLMService.InputParams] = None,
-        settings: Optional[GoogleLLMSettings] = None,
+        settings: Optional[GoogleVertexLLMSettings] = None,
         system_instruction: Optional[str] = None,
         tools: Optional[list] = None,
         tool_config: Optional[dict] = None,
@@ -183,7 +191,7 @@ class GoogleVertexLLMService(GoogleLLMService):
         self._location = location
 
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = GoogleLLMSettings(
+        default_settings = GoogleVertexLLMSettings(
             model="gemini-2.5-flash",
             max_tokens=4096,
             temperature=None,
@@ -200,12 +208,12 @@ class GoogleVertexLLMService(GoogleLLMService):
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", GoogleLLMSettings, "model")
+            _warn_deprecated_param("model", GoogleVertexLLMSettings, "model")
             default_settings.model = model
 
         # 3. Apply params overrides — only if settings not provided
         if params is not None:
-            _warn_deprecated_param("params", GoogleLLMSettings)
+            _warn_deprecated_param("params", GoogleVertexLLMSettings)
             if not settings:
                 default_settings.max_tokens = params.max_tokens
                 default_settings.temperature = params.temperature

src/pipecat/services/grok/llm.py

@@ -70,6 +70,13 @@ class GrokContextAggregatorPair:
         return self._assistant
 
 
+@dataclass
+class GrokLLMSettings(OpenAILLMSettings):
+    """Settings for Grok LLM service."""
+
+    pass
+
+
 class GrokLLMService(OpenAILLMService):
     """A service for interacting with Grok's API using the OpenAI-compatible interface.
 
@@ -85,7 +92,7 @@ class GrokLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://api.x.ai/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[GrokLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the GrokLLMService with API key and model.
@@ -103,11 +110,11 @@ class GrokLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="grok-3-beta")
+        default_settings = GrokLLMSettings(model="grok-3-beta")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", GrokLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/groq/llm.py

@@ -6,6 +6,7 @@
 
 """Groq LLM Service implementation using OpenAI-compatible interface."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -15,6 +16,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class GroqLLMSettings(OpenAILLMSettings):
+    """Settings for Groq LLM service."""
+
+    pass
+
+
 class GroqLLMService(OpenAILLMService):
     """A service for interacting with Groq's API using the OpenAI-compatible interface.
 
@@ -28,7 +36,7 @@ class GroqLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://api.groq.com/openai/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[GroqLLMSettings] = None,
         **kwargs,
     ):
         """Initialize Groq LLM service.
@@ -46,11 +54,11 @@ class GroqLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="llama-3.3-70b-versatile")
+        default_settings = GroqLLMSettings(model="llama-3.3-70b-versatile")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", GroqLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/mistral/llm.py

@@ -6,6 +6,7 @@
 
 """Mistral LLM service implementation using OpenAI-compatible interface."""
 
+from dataclasses import dataclass
 from typing import List, Optional, Sequence
 
 from loguru import logger
@@ -18,6 +19,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class MistralLLMSettings(OpenAILLMSettings):
+    """Settings for Mistral LLM service."""
+
+    pass
+
+
 class MistralLLMService(OpenAILLMService):
     """A service for interacting with Mistral's API using the OpenAI-compatible interface.
 
@@ -31,7 +39,7 @@ class MistralLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://api.mistral.ai/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[MistralLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the Mistral LLM service.
@@ -49,11 +57,11 @@ class MistralLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="mistral-small-latest")
+        default_settings = MistralLLMSettings(model="mistral-small-latest")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", MistralLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/nvidia/llm.py

@@ -10,6 +10,7 @@ This module provides a service for interacting with NVIDIA's NIM (NVIDIA Inferen
 Microservice) API while maintaining compatibility with the OpenAI-style interface.
 """
 
+from dataclasses import dataclass
 from typing import Optional
 
 from pipecat.metrics.metrics import LLMTokenUsage
@@ -20,6 +21,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class NvidiaLLMSettings(OpenAILLMSettings):
+    """Settings for NVIDIA LLM service."""
+
+    pass
+
+
 class NvidiaLLMService(OpenAILLMService):
     """A service for interacting with NVIDIA's NIM (NVIDIA Inference Microservice) API.
 
@@ -34,7 +42,7 @@ class NvidiaLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://integrate.api.nvidia.com/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[NvidiaLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the NvidiaLLMService.
@@ -53,11 +61,11 @@ class NvidiaLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="nvidia/llama-3.1-nemotron-70b-instruct")
+        default_settings = NvidiaLLMSettings(model="nvidia/llama-3.1-nemotron-70b-instruct")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", NvidiaLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/nvidia/stt.py

@@ -8,7 +8,7 @@
 
 import asyncio
 from concurrent.futures import CancelledError as FuturesCancelledError
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Any, AsyncGenerator, List, Mapping, Optional
 
 from loguru import logger
@@ -23,7 +23,7 @@ from pipecat.frames.frames import (
     StartFrame,
     TranscriptionFrame,
 )
-from pipecat.services.settings import STTSettings, _warn_deprecated_param
+from pipecat.services.settings import NOT_GIVEN, STTSettings, _NotGiven, _warn_deprecated_param
 from pipecat.services.stt_latency import NVIDIA_TTFS_P99
 from pipecat.services.stt_service import SegmentedSTTService, STTService
 from pipecat.transcriptions.language import Language, resolve_language
@@ -110,11 +110,11 @@ class NvidiaSegmentedSTTSettings(STTSettings):
         boosted_lm_score: Score boost for specified words.
     """
 
-    profanity_filter: bool = False
+    profanity_filter: bool | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
-    automatic_punctuation: bool = True
+    automatic_punctuation: bool | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
-    verbatim_transcripts: bool = False
+    verbatim_transcripts: bool | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
-    boosted_lm_words: Optional[List[str]] = None
+    boosted_lm_words: List[str] | None | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
-    boosted_lm_score: float = 4.0
+    boosted_lm_score: float | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
 
 
 class NvidiaSTTService(STTService):

src/pipecat/services/ollama/llm.py

@@ -6,6 +6,7 @@
 
 """OLLama LLM service implementation for Pipecat AI framework."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -15,6 +16,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class OllamaLLMSettings(OpenAILLMSettings):
+    """Settings for Ollama LLM service."""
+
+    pass
+
+
 class OLLamaLLMService(OpenAILLMService):
     """OLLama LLM service that provides local language model capabilities.
 
@@ -27,7 +35,7 @@ class OLLamaLLMService(OpenAILLMService):
         *,
         model: Optional[str] = None,
         base_url: str = "http://localhost:11434/v1",
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[OllamaLLMSettings] = None,
         **kwargs,
     ):
         """Initialize OLLama LLM service.
@@ -45,11 +53,11 @@ class OLLamaLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="llama2")
+        default_settings = OllamaLLMSettings(model="llama2")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", OllamaLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/openai/stt.py

@@ -16,7 +16,7 @@ Provides two STT services:
 
 import base64
 import json
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import Any, AsyncGenerator, Literal, Optional, Union
 
 from loguru import logger
@@ -35,7 +35,7 @@ from pipecat.frames.frames import (
     VADUserStoppedSpeakingFrame,
 )
 from pipecat.processors.frame_processor import FrameDirection
-from pipecat.services.settings import STTSettings, _NotGiven, _warn_deprecated_param
+from pipecat.services.settings import NOT_GIVEN, STTSettings, _NotGiven, _warn_deprecated_param
 from pipecat.services.stt_latency import OPENAI_REALTIME_TTFS_P99, OPENAI_TTFS_P99
 from pipecat.services.stt_service import WebsocketSTTService
 from pipecat.services.whisper.base_stt import (
@@ -188,7 +188,7 @@ class OpenAIRealtimeSTTSettings(STTSettings):
         prompt: Optional prompt text to guide transcription style.
     """
 
-    prompt: str | None | _NotGiven = None
+    prompt: str | None | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
 
 
 class OpenAIRealtimeSTTService(WebsocketSTTService):

src/pipecat/services/openai_realtime_beta/azure.py

@@ -7,10 +7,11 @@
 """Azure OpenAI Realtime Beta LLM service implementation."""
 
 import warnings
+from dataclasses import dataclass
 
 from loguru import logger
 
-from .openai import OpenAIRealtimeBetaLLMService
+from .openai import OpenAIRealtimeBetaLLMService, OpenAIRealtimeBetaLLMSettings
 
 try:
     from websockets.asyncio.client import connect as websocket_connect
@@ -22,6 +23,13 @@ except ModuleNotFoundError as e:
     raise Exception(f"Missing module: {e}")
 
 
+@dataclass
+class AzureRealtimeBetaLLMSettings(OpenAIRealtimeBetaLLMSettings):
+    """Settings for Azure Realtime Beta LLM service."""
+
+    pass
+
+
 class AzureRealtimeBetaLLMService(OpenAIRealtimeBetaLLMService):
     """Azure OpenAI Realtime Beta LLM service with Azure-specific authentication.
 

src/pipecat/services/openpipe/llm.py

@@ -10,6 +10,7 @@ This module provides an OpenPipe-specific implementation of the OpenAI LLM servi
 enabling integration with OpenPipe's fine-tuning and monitoring capabilities.
 """
 
+from dataclasses import dataclass
 from typing import Dict, Optional
 
 from loguru import logger
@@ -27,6 +28,13 @@ except ModuleNotFoundError as e:
     raise Exception(f"Missing module: {e}")
 
 
+@dataclass
+class OpenPipeLLMSettings(OpenAILLMSettings):
+    """Settings for OpenPipe LLM service."""
+
+    pass
+
+
 class OpenPipeLLMService(OpenAILLMService):
     """OpenPipe-powered Large Language Model service.
 
@@ -44,7 +52,7 @@ class OpenPipeLLMService(OpenAILLMService):
         openpipe_api_key: Optional[str] = None,
         openpipe_base_url: str = "https://app.openpipe.ai/api/v1",
         tags: Optional[Dict[str, str]] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[OpenPipeLLMSettings] = None,
         **kwargs,
     ):
         """Initialize OpenPipe LLM service.
@@ -65,11 +73,11 @@ class OpenPipeLLMService(OpenAILLMService):
             **kwargs: Additional arguments passed to parent OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="gpt-4.1")
+        default_settings = OpenPipeLLMSettings(model="gpt-4.1")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", OpenPipeLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/openrouter/llm.py

@@ -10,6 +10,7 @@ This module provides an OpenAI-compatible interface for interacting with OpenRou
 extending the base OpenAI LLM service functionality.
 """
 
+from dataclasses import dataclass
 from typing import Any, Dict, Optional
 
 from loguru import logger
@@ -19,6 +20,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class OpenRouterLLMSettings(OpenAILLMSettings):
+    """Settings for OpenRouter LLM service."""
+
+    pass
+
+
 class OpenRouterLLMService(OpenAILLMService):
     """A service for interacting with OpenRouter's API using the OpenAI-compatible interface.
 
@@ -32,7 +40,7 @@ class OpenRouterLLMService(OpenAILLMService):
         api_key: Optional[str] = None,
         model: Optional[str] = None,
         base_url: str = "https://openrouter.ai/api/v1",
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[OpenRouterLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the OpenRouter LLM service.
@@ -51,11 +59,11 @@ class OpenRouterLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="openai/gpt-4o-2024-11-20")
+        default_settings = OpenRouterLLMSettings(model="openai/gpt-4o-2024-11-20")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", OpenRouterLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/perplexity/llm.py

@@ -11,6 +11,7 @@ an OpenAI-compatible interface. It handles Perplexity's unique token usage
 reporting patterns while maintaining compatibility with the Pipecat framework.
 """
 
+from dataclasses import dataclass
 from typing import Optional
 
 from pipecat.adapters.services.open_ai_adapter import OpenAILLMInvocationParams
@@ -22,6 +23,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class PerplexityLLMSettings(OpenAILLMSettings):
+    """Settings for Perplexity LLM service."""
+
+    pass
+
+
 class PerplexityLLMService(OpenAILLMService):
     """A service for interacting with Perplexity's API.
 
@@ -36,7 +44,7 @@ class PerplexityLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://api.perplexity.ai",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[PerplexityLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the Perplexity LLM service.
@@ -54,11 +62,11 @@ class PerplexityLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="sonar")
+        default_settings = PerplexityLLMSettings(model="sonar")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", PerplexityLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/qwen/llm.py

@@ -6,6 +6,7 @@
 
 """Qwen LLM service implementation using OpenAI-compatible interface."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -15,6 +16,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class QwenLLMSettings(OpenAILLMSettings):
+    """Settings for Qwen LLM service."""
+
+    pass
+
+
 class QwenLLMService(OpenAILLMService):
     """A service for interacting with Alibaba Cloud's Qwen LLM API using the OpenAI-compatible interface.
 
@@ -28,7 +36,7 @@ class QwenLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[QwenLLMSettings] = None,
         **kwargs,
     ):
         """Initialize the Qwen LLM service.
@@ -46,11 +54,11 @@ class QwenLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="qwen-plus")
+        default_settings = QwenLLMSettings(model="qwen-plus")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", QwenLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/sambanova/llm.py

@@ -7,6 +7,7 @@
 """SambaNova LLM service implementation using OpenAI-compatible interface."""
 
 import json
+from dataclasses import dataclass
 from typing import Any, Dict, Optional
 
 from loguru import logger
@@ -27,6 +28,13 @@ from pipecat.services.settings import _warn_deprecated_param
 from pipecat.utils.tracing.service_decorators import traced_llm
 
 
+@dataclass
+class SambaNovaLLMSettings(OpenAILLMSettings):
+    """Settings for SambaNova LLM service."""
+
+    pass
+
+
 class SambaNovaLLMService(OpenAILLMService):  # type: ignore
     """A service for interacting with SambaNova using the OpenAI-compatible interface.
 
@@ -40,7 +48,7 @@ class SambaNovaLLMService(OpenAILLMService):  # type: ignore
         api_key: str,
         model: Optional[str] = None,
         base_url: str = "https://api.sambanova.ai/v1",
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[SambaNovaLLMSettings] = None,
         **kwargs: Dict[Any, Any],
     ) -> None:
         """Initialize SambaNova LLM service.
@@ -58,11 +66,11 @@ class SambaNovaLLMService(OpenAILLMService):  # type: ignore
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="Llama-4-Maverick-17B-128E-Instruct")
+        default_settings = SambaNovaLLMSettings(model="Llama-4-Maverick-17B-128E-Instruct")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", SambaNovaLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/together/llm.py

@@ -6,6 +6,7 @@
 
 """Together.ai LLM service implementation using OpenAI-compatible interface."""
 
+from dataclasses import dataclass
 from typing import Optional
 
 from loguru import logger
@@ -15,6 +16,13 @@ from pipecat.services.openai.llm import OpenAILLMService
 from pipecat.services.settings import _warn_deprecated_param
 
 
+@dataclass
+class TogetherLLMSettings(OpenAILLMSettings):
+    """Settings for Together LLM service."""
+
+    pass
+
+
 class TogetherLLMService(OpenAILLMService):
     """A service for interacting with Together.ai's API using the OpenAI-compatible interface.
 
@@ -28,7 +36,7 @@ class TogetherLLMService(OpenAILLMService):
         api_key: str,
         base_url: str = "https://api.together.xyz/v1",
         model: Optional[str] = None,
-        settings: Optional[OpenAILLMSettings] = None,
+        settings: Optional[TogetherLLMSettings] = None,
         **kwargs,
     ):
         """Initialize Together.ai LLM service.
@@ -46,11 +54,11 @@ class TogetherLLMService(OpenAILLMService):
             **kwargs: Additional keyword arguments passed to OpenAILLMService.
         """
         # 1. Initialize default_settings with hardcoded defaults
-        default_settings = OpenAILLMSettings(model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo")
+        default_settings = TogetherLLMSettings(model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo")
 
         # 2. Apply direct init arg overrides (deprecated)
         if model is not None:
-            _warn_deprecated_param("model", OpenAILLMSettings, "model")
+            _warn_deprecated_param("model", TogetherLLMSettings, "model")
             default_settings.model = model
 
         # 4. Apply settings delta (canonical API, always wins)

src/pipecat/services/whisper/base_stt.py

@@ -10,7 +10,7 @@ This module provides common functionality for services implementing the Whisper
 interface, including language mapping, metrics generation, and error handling.
 """
 
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from typing import AsyncGenerator, Optional
 
 from loguru import logger
@@ -18,7 +18,7 @@ from openai import AsyncOpenAI
 from openai.types.audio import Transcription
 
 from pipecat.frames.frames import ErrorFrame, Frame, TranscriptionFrame
-from pipecat.services.settings import STTSettings, _NotGiven, _warn_deprecated_param
+from pipecat.services.settings import NOT_GIVEN, STTSettings, _NotGiven, _warn_deprecated_param
 from pipecat.services.stt_latency import WHISPER_TTFS_P99
 from pipecat.services.stt_service import SegmentedSTTService
 from pipecat.transcriptions.language import Language, resolve_language
@@ -36,8 +36,8 @@ class BaseWhisperSTTSettings(STTSettings):
         temperature: Sampling temperature between 0 and 1.
     """
 
-    prompt: str | None | _NotGiven = None
+    prompt: str | None | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
-    temperature: float | None | _NotGiven = None
+    temperature: float | None | _NotGiven = field(default_factory=lambda: NOT_GIVEN)
 
 
 def language_to_whisper_language(language: Language) -> Optional[str]:
@@ -183,6 +183,8 @@ class BaseWhisperSTTService(SegmentedSTTService):
         default_settings = BaseWhisperSTTSettings(
             model=None,
             language=None,
+            prompt=None,
+            temperature=None,
         )
 
         # --- 2. Deprecated direct-arg overrides ---

tests/test_service_init.py

@@ -0,0 +1,181 @@
+#
+# Copyright (c) 2024-2026, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+"""Tests for service settings and initialization patterns.
+
+Settings objects operate in two modes:
+
+- **Store mode** (``self._settings``): the live state inside a service.
+  Every field must hold a real value (``None`` is fine, ``NOT_GIVEN`` is not).
+- **Delta mode** (``FooSettings()`` with no args): a sparse update.
+  Every field must default to ``NOT_GIVEN`` so ``apply_update()`` skips
+  untouched fields and doesn't accidentally overwrite the store.
+
+These tests verify both sides of that contract automatically:
+
+1. **Delta defaults** — Instantiate every ``ServiceSettings`` subclass with
+   no arguments and assert that every field is ``NOT_GIVEN``.  Catches the
+   bug where a field defaults to ``None`` instead of ``NOT_GIVEN``, which
+   would cause partial deltas to silently overwrite unrelated store values.
+
+2. **Store completeness** — Instantiate every concrete service with dummy
+   args and assert that ``_settings`` contains no ``NOT_GIVEN`` values.
+   This is the same check that ``validate_complete()`` runs in ``start()``,
+   but caught here at unit-test time without needing a running pipeline.
+   Catches services that forget to initialize a field in ``default_settings``.
+
+All Settings and Service classes are auto-discovered via ``pkgutil``;
+new services are covered automatically with no per-service maintenance.
+"""
+
+import importlib
+import inspect
+import pkgutil
+import warnings
+from dataclasses import fields
+
+import pytest
+
+import pipecat.services
+from pipecat.services.ai_service import AIService
+from pipecat.services.settings import ServiceSettings, is_given
+
+# Modules that define abstract base service classes (not concrete services).
+_BASE_MODULES = frozenset(
+    {
+        "pipecat.services.ai_service",
+        "pipecat.services.llm_service",
+        "pipecat.services.stt_service",
+        "pipecat.services.tts_service",
+        "pipecat.services.image_gen_service",
+        "pipecat.services.vision_service",
+    }
+)
+
+
+# ---------------------------------------------------------------------------
+# Auto-discovery
+# ---------------------------------------------------------------------------
+
+
+def _all_subclasses(cls):
+    result = set()
+    for sub in cls.__subclasses__():
+        result.add(sub)
+        result.update(_all_subclasses(sub))
+    return result
+
+
+def _import_all_service_modules():
+    """Import every module under pipecat.services (skipping missing deps)."""
+    package = pipecat.services
+    for _importer, modname, _ispkg in pkgutil.walk_packages(
+        package.__path__, prefix=package.__name__ + "."
+    ):
+        try:
+            importlib.import_module(modname)
+        except Exception:
+            continue
+
+
+_import_all_service_modules()
+
+ALL_SETTINGS_CLASSES = sorted(_all_subclasses(ServiceSettings), key=lambda c: c.__qualname__)
+assert ALL_SETTINGS_CLASSES, "No settings classes discovered"
+
+
+# ---------------------------------------------------------------------------
+# Service instantiation helpers
+# ---------------------------------------------------------------------------
+
+
+def _try_instantiate(cls):
+    """Try to instantiate a service with dummy values for required args.
+
+    Inspects the __init__ signature and passes "test" for every required
+    keyword-only parameter.  Services that need non-string required args
+    or fail for other reasons will raise and be skipped by the test.
+    """
+    sig = inspect.signature(cls.__init__)
+    kwargs = {}
+    for name, param in sig.parameters.items():
+        if name == "self":
+            continue
+        if param.kind in (param.VAR_POSITIONAL, param.VAR_KEYWORD):
+            continue
+        if param.default is not param.empty:
+            continue
+        # Required parameter — pass a dummy string
+        kwargs[name] = "test"
+    return cls(**kwargs)
+
+
+def _discover_service_classes():
+    """Return concrete service classes that can be instantiated with dummy args."""
+    result = []
+    for cls in sorted(_all_subclasses(AIService), key=lambda c: c.__qualname__):
+        # Skip abstract base classes defined in framework modules.
+        if cls.__module__ in _BASE_MODULES:
+            continue
+        try:
+            svc = _try_instantiate(cls)
+        except Exception:
+            continue
+        if hasattr(svc, "_settings"):
+            result.append(cls)
+    return result
+
+
+ALL_SERVICE_CLASSES = _discover_service_classes()
+assert ALL_SERVICE_CLASSES, "No service classes could be instantiated"
+
+
+# ---------------------------------------------------------------------------
+# 1. Settings defaults: delta-mode safety
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize("settings_cls", ALL_SETTINGS_CLASSES, ids=lambda c: c.__qualname__)
+def test_delta_defaults_are_not_given(settings_cls):
+    """Every field must default to NOT_GIVEN so empty deltas are no-ops.
+
+    A field that defaults to None instead of NOT_GIVEN will cause
+    apply_update() to overwrite the corresponding store value whenever
+    a partial delta is applied.
+    """
+    instance = settings_cls()
+    for f in fields(instance):
+        if f.name == "extra":
+            continue
+        val = getattr(instance, f.name)
+        assert not is_given(val), (
+            f"{settings_cls.__qualname__}.{f.name} defaults to {val!r}, expected NOT_GIVEN"
+        )
+
+
+# ---------------------------------------------------------------------------
+# 2. Service construction: store-mode completeness
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize("service_cls", ALL_SERVICE_CLASSES, ids=lambda c: c.__qualname__)
+def test_service_settings_complete(service_cls):
+    """After construction, _settings must have no NOT_GIVEN values.
+
+    This is what validate_complete() checks in start().  Catching it
+    here means we don't need a running pipeline to find missing defaults.
+    """
+    try:
+        svc = _try_instantiate(service_cls)
+    except Exception:
+        pytest.skip("Cannot re-instantiate (environment issue)")
+    for f in fields(svc._settings):
+        if f.name == "extra":
+            continue
+        val = getattr(svc._settings, f.name)
+        assert is_given(val), (
+            f"{service_cls.__qualname__}._settings.{f.name} is NOT_GIVEN after construction"
+        )