feat(provider): Add Google GCP support

backend/app/core/cloud/storage.py

@@ -307,7 +307,7 @@ def get_cloud_storage(session: Session, project_id: int) -> CloudStorage:
     Method to create and configure a cloud storage instance.
     """
     # Lazy import to avoid a top-level cycle: storage.py is imported from
-    # app.services.llm.providers.google_ai, which itself is wired into the
+    # app.services.llm.providers.google_gcp, which itself is wired into the
     # provider registry that app.crud transitively pulls in.
     from app.crud import get_project_by_id
 

backend/app/core/providers.py

@@ -12,6 +12,7 @@ class Provider(str, Enum):
     OPENAI = "openai"
     LANGFUSE = "langfuse"
     GOOGLE_AISTUDIO = "google-aistudio"
+    GOOGLE_GCP = "google-gcp"
     SARVAMAI = "sarvamai"
     ELEVENLABS = "elevenlabs"
     ANTHROPIC = "anthropic"
@@ -55,6 +56,16 @@ class ProviderConfig:
         ],
         sensitive_fields=["api_key"],
     ),
+    Provider.GOOGLE_GCP: ProviderConfig(
+        required_fields=[
+            "api_key",
+            "project_id",
+            "location",
+            "sa_key",
+            "gcs_bucket",
+        ],
+        sensitive_fields=["api_key", "sa_key"],
+    ),
     Provider.WEBHOOK_SECRET: ProviderConfig(
         required_fields=["webhook_secret"], sensitive_fields=["webhook_secret"]
     ),

backend/app/crud/credentials.py

@@ -199,24 +199,34 @@ def update_creds_for_org(
     ):
         credential_data = credential_data[creds_in.provider]
 
+    statement = select(Credential).where(
+        Credential.organization_id == org_id,
+        Credential.provider == creds_in.provider,
+        Credential.is_active.is_(True),
+        Credential.project_id == project_id,
+    )
+    creds = session.exec(statement).one_or_none()
+
+    # Merge onto the existing credentials so a partial payload (PATCH) only
+    # overwrites the fields it supplies instead of dropping the rest.
+    merged_credential_data = credential_data
+    if creds and creds.credential:
+        merged_credential_data = {
+            **decrypt_credentials(creds.credential),
+            **credential_data,
+        }
+
     try:
-        validate_provider_credentials(creds_in.provider, credential_data)
+        validate_provider_credentials(creds_in.provider, merged_credential_data)
     except ValueError as e:
         logger.warning(
             f"[update_creds_for_org] Validation error | organization_id: {org_id}, project_id: {project_id}, provider: {creds_in.provider}, error: {str(e)}"
         )
         raise HTTPException(status_code=400, detail=str(e))
 
     # Encrypt the entire credentials object
-    encrypted_credentials = encrypt_credentials(credential_data)
+    encrypted_credentials = encrypt_credentials(merged_credential_data)
 
-    statement = select(Credential).where(
-        Credential.organization_id == org_id,
-        Credential.provider == creds_in.provider,
-        Credential.is_active.is_(True),
-        Credential.project_id == project_id,
-    )
-    creds = session.exec(statement).one_or_none()
     if creds is None:
         # Create new credential if it doesn't exist
         creds = Credential(

backend/app/models/llm/constants.py

@@ -9,6 +9,7 @@ class Provider(StrEnum):
     ELEVENLABS = "elevenlabs"
     ANTHROPIC = "anthropic"
     GOOGLE_AISTUDIO = "google-aistudio"
+    GOOGLE_GCP = "google-gcp"
     PROXY = "proxy"
 
 
@@ -17,12 +18,14 @@ class Provider(StrEnum):
 # instead of leaving behind stale magic strings.
 STTProvider = Literal[
     Provider.GOOGLE,
+    Provider.GOOGLE_GCP,
     Provider.SARVAMAI,
     Provider.ELEVENLABS,
     Provider.GOOGLE_AISTUDIO,
 ]
 TTSProvider = Literal[
     Provider.GOOGLE,
+    Provider.GOOGLE_GCP,
     Provider.SARVAMAI,
     Provider.ELEVENLABS,
     Provider.GOOGLE_AISTUDIO,
@@ -32,6 +35,7 @@ class Provider(StrEnum):
 KaapiProvider = Literal[
     Provider.OPENAI,
     Provider.GOOGLE,
+    Provider.GOOGLE_GCP,
     Provider.SARVAMAI,
     Provider.ELEVENLABS,
     Provider.ANTHROPIC,
@@ -49,6 +53,7 @@ class Provider(StrEnum):
     "elevenlabs-native",
     "anthropic-native",
     "google-aistudio-native",
+    "google-gcp-native",
 ]
 
 

backend/app/models/llm/request.py

@@ -295,8 +295,8 @@ class KaapiCompletionConfig(SQLModel):
         None,
         description=(
             "LLM provider (openai, google, sarvamai, elevenlabs, anthropic, "
-            "google-aistudio). 'google' routes via Google Vertex AI; "
-            "'google-aistudio' uses Google AI Studio."
+            "google-aistudio, google-gcp). 'google-aistudio' uses Google AI "
+            "Studio; 'google-gcp' uses Google GCP directly for STT/TTS."
         ),
     )
 

backend/app/models/model_config.py

@@ -18,6 +18,7 @@ class ModelConfigBase(SQLModel):
         "anthropic",
         "proxy",
         "google-aistudio",
+        "google-gcp",
     ] = Field(
         default="openai",
         sa_column=sa.Column(
@@ -29,12 +30,13 @@ class ModelConfigBase(SQLModel):
                 "anthropic",
                 "proxy",
                 "google-aistudio",
+                "google-gcp",
                 name="provider_enum",
                 schema="global",
                 create_type=False,
             ),
             nullable=False,
-            comment="provider name (e.g. openai, google, sarvamai, elevenlabs, anthropic, google-aistudio, proxy)",
+            comment="provider name (e.g. openai, google, sarvamai, elevenlabs, anthropic, google-aistudio, google-gcp, proxy)",
         ),
     )
 

backend/app/services/llm/mappers.py

@@ -696,6 +696,20 @@ def transform_kaapi_config_to_native(
             warnings,
         )
 
+    if kaapi_config.provider == Provider.GOOGLE_GCP:
+        # Kaapi STT/TTS param shape is identical to Google's; reuse the Google mapper.
+        mapped_params, warnings = map_kaapi_to_google_params(
+            kaapi_config.params, kaapi_config.type
+        )
+        return (
+            NativeCompletionConfig(
+                provider="google-gcp-native",
+                params=mapped_params,
+                type=kaapi_config.type,
+            ),
+            warnings,
+        )
+
     if kaapi_config.provider == Provider.ANTHROPIC:
         if kaapi_config.type != CompletionType.TEXT:
             raise ValueError(

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

@@ -4,7 +4,7 @@
 from app.services.llm.providers.eleven_ai import ElevenlabsAIProvider
 from app.services.llm.providers.sarvam_ai import SarvamAIProvider
 from app.services.llm.providers.claude import ClaudeProvider
-from app.services.llm.providers.google_ai import GoogleVertexAIProvider
+from app.services.llm.providers.google_gcp import GoogleGCPProvider
 from app.services.llm.providers.registry import (
     LLMProvider,
     get_llm_provider,