Gemini: Toggleable Inference Routing

backend/app/core/config.py

@@ -115,6 +115,8 @@ def SQLALCHEMY_DATABASE_URI(self) -> PostgresDsn:
     GCP_SA_KEY: str = ""
     GCS_AUDIO_BUCKET: str = ""
 
+    GEMINI_API_KEY: str = ""
+    GEMINI_DEFAULT_INFERENCE_ROUTE: str = "aistudio"
     # RabbitMQ configuration for Celery broker
     RABBITMQ_HOST: str = "localhost"
     RABBITMQ_PORT: int = 5672

backend/app/models/llm/constants.py

@@ -9,6 +9,7 @@ class Provider(StrEnum):
     ELEVENLABS = "elevenlabs"
     ANTHROPIC = "anthropic"
     GOOGLE_AISTUDIO = "google-aistudio"
+    GOOGLE_VERTEX = "google-vertex"
     PROXY = "proxy"
 
 
@@ -47,6 +48,7 @@ class Provider(StrEnum):
     "elevenlabs-native",
     "anthropic-native",
     "google-aistudio-native",
+    "google-vertex-native",
 ]
 
 

backend/app/services/llm/providers/google_ai.py

@@ -94,7 +94,12 @@ def __init__(
         self.gcs_bucket = gcs_bucket or settings.GCS_AUDIO_BUCKET
 
     def endpoint(self, model: str) -> str:
-        host = N
+        host = ""
+        if self.location == "global":
+            host = "aiplatform.googleapis.com"
+        else:
+            host = f"{self.location}-aiplatform.googleapis.com"
+
         return (
             f"https://{host}/v1"
             f"/projects/{self.project_id}/locations/{self.location}"

backend/app/services/llm/providers/google_aistudio.py

@@ -33,6 +33,7 @@
 from app.models.llm.response import AudioOutput, AudioContent
 from app.services.llm.providers.base import BaseProvider, ContentPart, MultiModalInput
 from app.services.llm.mappers import BCP47_LOCALE_TO_GEMINI_LANG
+from app.core.config import settings
 from app.core.audio_utils import (
     AudioRef,
     convert_pcm_to_mp3,
@@ -55,9 +56,10 @@ def __init__(self, client: genai.Client):
 
     @staticmethod
     def create_client(credentials: dict[str, Any]) -> Any:
-        if "api_key" not in credentials:
+        api_key = credentials.get("api_key") or settings.GEMINI_API_KEY
+        if not api_key:
             raise ValueError("API Key for Google Gemini Not Set")
-        return genai.Client(api_key=credentials["api_key"])
+        return genai.Client(api_key=api_key)
 
     @staticmethod
     def format_parts(

backend/app/services/llm/providers/registry.py

@@ -1,9 +1,11 @@
 import logging
 from sqlmodel import Session
 
+from app.core.config import settings
 from app.services.llm.providers.base import BaseProvider
 from app.services.llm.providers.open_ai import OpenAIProvider
 from app.services.llm.providers.google_aistudio import GoogleAIProvider
+from app.services.llm.providers.google_ai import GoogleVertexAIProvider
 from app.services.llm.providers.sarvam_ai import SarvamAIProvider
 from app.services.llm.providers.eleven_ai import ElevenlabsAIProvider
 from app.services.llm.providers.claude import ClaudeProvider
@@ -15,34 +17,46 @@ class LLMProvider:
     OPENAI = "openai"
     SARVAMAI = "sarvamai"
     ELEVENLABS = "elevenlabs"
-    GOOGLE = "google"
     ANTHROPIC = "anthropic"
+
+    # Platform-routed Google. Resolved via GEMINI_DEFAULT_INFERENCE_ROUTE.
+    GOOGLE = "google"
+    GOOGLE_NATIVE = "google-native"
+    # Explicit Google backends. Bypass env routing; use caller's own creds.
+    GOOGLE_VERTEX = "google-vertex"
+    GOOGLE_VERTEX_NATIVE = "google-vertex-native"
     GOOGLE_AISTUDIO = "google-aistudio"
     GOOGLE_AISTUDIO_NATIVE = "google-aistudio-native"
+
     OPENAI_NATIVE = "openai-native"
-    GOOGLE_NATIVE = "google-native"
     SARVAMAI_NATIVE = "sarvamai-native"
     ELEVENLABS_NATIVE = "elevenlabs-native"
     ANTHROPIC_NATIVE = "anthropic-native"
 
     _registry: dict[str, type[BaseProvider]] = {
         OPENAI: OpenAIProvider,
-        GOOGLE: GoogleAIProvider,
         SARVAMAI: SarvamAIProvider,
         ELEVENLABS: ElevenlabsAIProvider,
         ANTHROPIC: ClaudeProvider,
+        GOOGLE: GoogleVertexAIProvider,  # placeholder; env decides at resolve time
+        GOOGLE_NATIVE: GoogleVertexAIProvider,
+        GOOGLE_VERTEX: GoogleVertexAIProvider,
+        GOOGLE_VERTEX_NATIVE: GoogleVertexAIProvider,
         GOOGLE_AISTUDIO: GoogleAIProvider,
         GOOGLE_AISTUDIO_NATIVE: GoogleAIProvider,
         OPENAI_NATIVE: OpenAIProvider,
-        GOOGLE_NATIVE: GoogleAIProvider,
         SARVAMAI_NATIVE: SarvamAIProvider,
         ELEVENLABS_NATIVE: ElevenlabsAIProvider,
         ANTHROPIC_NATIVE: ClaudeProvider,
     }
 
+    _GOOGLE_ROUTED = {GOOGLE, GOOGLE_NATIVE}
+
     @classmethod
     def get_provider_class(cls, provider_type: str) -> type[BaseProvider]:
-        """Return the provider class for a given name."""
+        if provider_type in cls._GOOGLE_ROUTED:
+            route = settings.GEMINI_DEFAULT_INFERENCE_ROUTE
+            return GoogleAIProvider if route == "aistudio" else GoogleVertexAIProvider
         provider = cls._registry.get(provider_type)
         if not provider:
             raise ValueError(
@@ -53,7 +67,6 @@ def get_provider_class(cls, provider_type: str) -> type[BaseProvider]:
 
     @classmethod
     def supported_providers(cls) -> list[str]:
-        """Return a list of supported provider names."""
         return list(cls._registry.keys())
 
 
@@ -63,8 +76,16 @@ def get_llm_provider(
     from app.crud.credentials import get_provider_credential
 
     provider_class = LLMProvider.get_provider_class(provider_type)
+    is_platform_routed = provider_type in LLMProvider._GOOGLE_ROUTED
 
-    credential_provider = provider_type.replace("-native", "")
+    if is_platform_routed:
+        credential_provider = (
+            LLMProvider.GOOGLE_AISTUDIO
+            if provider_class is GoogleAIProvider
+            else LLMProvider.GOOGLE_VERTEX
+        )
+    else:
+        credential_provider = provider_type.replace("-native", "")
 
     credentials = get_provider_credential(
         session=session,
@@ -73,16 +94,17 @@ def get_llm_provider(
         org_id=organization_id,
     )
 
-    if not credentials:
+    if not credentials and not is_platform_routed:
         raise ValueError(
             f"Credentials for provider '{credential_provider}' not configured for this project."
         )
 
+    credentials = credentials or {}
+
     try:
         client = provider_class.create_client(credentials=credentials)
         return provider_class(client=client)
     except ValueError:
-        # Re-raise ValueError for credential/configuration errors
         raise
     except Exception as e:
         logger.error(f"Failed to initialize {provider_type} client: {e}", exc_info=True)