fix(provider): support custom Azure deployments

flocks/server/routes/provider.py

@@ -2320,9 +2320,15 @@ async def test_provider_credentials(provider_id: str, body: Optional[TestCredent
             requested_model_id = body.model_id if body else None
             test_model_id = requested_model_id or models[0].id
 
-            # Validate model belongs to this provider
+            # Validate model belongs to this provider. Azure OpenAI is the
+            # exception: users may test a deployment name before saving it.
             valid_ids = {m.id for m in models}
-            if test_model_id not in valid_ids:
+            is_unsaved_azure_deployment = (
+                requested_model_id
+                and provider_id in {"azure-openai", "azure"}
+                and test_model_id not in valid_ids
+            )
+            if test_model_id not in valid_ids and not is_unsaved_azure_deployment:
                 response = {
                     "success": False,
                     "message": f"模型 '{test_model_id}' 不属于该 Provider",

tests/provider/test_azure_provider.py

@@ -0,0 +1,33 @@
+from flocks.provider.provider import ModelCapabilities, ModelInfo
+from flocks.provider.sdk.azure import AzureProvider
+
+
+def test_azure_provider_returns_configured_deployment_models():
+    provider = AzureProvider()
+    provider._config_models = [
+        ModelInfo(
+            id="customer-prod-deployment",
+            name="Customer Production Deployment",
+            provider_id="azure",
+            capabilities=ModelCapabilities(
+                supports_tools=True,
+                supports_streaming=True,
+                context_window=128000,
+                max_tokens=4096,
+            ),
+        )
+    ]
+
+    models = provider.get_models()
+
+    assert [m.id for m in models] == ["customer-prod-deployment"]
+    assert models[0].name == "Customer Production Deployment"
+
+
+def test_azure_provider_returns_fallback_models_without_config():
+    provider = AzureProvider()
+
+    models = provider.get_models()
+
+    assert {m.id for m in models} == {"gpt-5.4", "gpt-5-mini"}
+    assert all(m.provider_id == "azure" for m in models)

tests/provider/test_test_credentials.py

@@ -754,3 +754,83 @@ async def test_existing_custom_settings_are_preserved_during_provider_test(self)
             assert configured.api_key == "gateway-api-key"
             assert configured.base_url == "https://gateway.internal/v1"
             assert configured.custom_settings["verify_ssl"] is False
+
+    @pytest.mark.asyncio
+    async def test_requested_azure_deployment_model_is_used_for_provider_test(self):
+        from flocks.server.routes.provider import TestCredentialRequest, test_provider_credentials
+
+        provider = MagicMock()
+        provider._config = MagicMock(
+            custom_settings={},
+            base_url="https://example-resource.openai.azure.com/",
+        )
+        provider.chat = AsyncMock(return_value=MagicMock(content="Paris"))
+
+        model = MagicMock()
+        model.id = "customer-prod-deployment"
+
+        mock_secrets = MagicMock()
+        mock_secrets.get.return_value = "azure-api-key"
+
+        mock_config = MagicMock()
+
+        with (
+            patch(_PATCH_SECRET_MGR, return_value=mock_secrets),
+            patch(_PATCH_CONFIG_GET, new_callable=AsyncMock, return_value=mock_config),
+            patch(_PATCH_PROVIDER) as mock_provider_cls,
+        ):
+            mock_provider_cls._ensure_initialized = MagicMock()
+            mock_provider_cls._load_dynamic_providers = MagicMock()
+            mock_provider_cls.apply_config = AsyncMock()
+            mock_provider_cls.get.return_value = provider
+            mock_provider_cls.list_models.return_value = [model]
+
+            result = await test_provider_credentials(
+                "azure-openai",
+                TestCredentialRequest(model_id="customer-prod-deployment"),
+            )
+
+            assert result["success"] is True, result
+            assert result["model_id"] == "customer-prod-deployment"
+            provider.chat.assert_awaited_once()
+            assert provider.chat.await_args.args[0] == "customer-prod-deployment"
+
+    @pytest.mark.asyncio
+    async def test_unsaved_azure_deployment_can_be_tested_without_model_definition(self):
+        from flocks.server.routes.provider import TestCredentialRequest, test_provider_credentials
+
+        provider = MagicMock()
+        provider._config = MagicMock(
+            custom_settings={},
+            base_url="https://example-resource.openai.azure.com/",
+        )
+        provider.chat = AsyncMock(return_value=MagicMock(content="Paris"))
+
+        catalog_model = MagicMock()
+        catalog_model.id = "gpt-5.4"
+
+        mock_secrets = MagicMock()
+        mock_secrets.get.return_value = "azure-api-key"
+
+        mock_config = MagicMock()
+
+        with (
+            patch(_PATCH_SECRET_MGR, return_value=mock_secrets),
+            patch(_PATCH_CONFIG_GET, new_callable=AsyncMock, return_value=mock_config),
+            patch(_PATCH_PROVIDER) as mock_provider_cls,
+        ):
+            mock_provider_cls._ensure_initialized = MagicMock()
+            mock_provider_cls._load_dynamic_providers = MagicMock()
+            mock_provider_cls.apply_config = AsyncMock()
+            mock_provider_cls.get.return_value = provider
+            mock_provider_cls.list_models.return_value = [catalog_model]
+
+            result = await test_provider_credentials(
+                "azure-openai",
+                TestCredentialRequest(model_id="unsaved-prod-deployment"),
+            )
+
+            assert result["success"] is True, result
+            assert result["model_id"] == "unsaved-prod-deployment"
+            provider.chat.assert_awaited_once()
+            assert provider.chat.await_args.args[0] == "unsaved-prod-deployment"

tests/server/routes/test_custom_provider_runtime.py

@@ -1,4 +1,5 @@
 from flocks.provider.provider import ModelCapabilities, ModelInfo, Provider
+from flocks.provider.sdk.azure import AzureProvider
 from flocks.server.routes.custom_provider import CreateModelReq, _add_model_to_runtime
 
 
@@ -48,3 +49,35 @@ class DummyProvider:
         assert provider._config_models[0].capabilities.supports_reasoning is True
     finally:
         Provider._models = original_models
+
+
+def test_add_azure_deployment_to_runtime_config_models(monkeypatch):
+    provider = AzureProvider()
+    provider.id = "azure-openai"
+    provider._config_models = []
+    body = CreateModelReq(
+        model_id="customer-prod-deployment",
+        name="Customer Production Deployment",
+        context_window=128000,
+        max_output_tokens=4096,
+        supports_vision=False,
+        supports_tools=True,
+        supports_streaming=True,
+        supports_reasoning=False,
+        input_price=0.0,
+        output_price=0.0,
+        currency="USD",
+    )
+
+    original_models = Provider._models
+    Provider._models = {}
+    monkeypatch.setattr(Provider, "get", classmethod(lambda cls, provider_id: provider))
+
+    try:
+        _add_model_to_runtime("azure-openai", body)
+
+        assert Provider._models[body.model_id].provider_id == "azure-openai"
+        assert provider._config_models[0].id == "customer-prod-deployment"
+        assert provider._config_models[0].name == "Customer Production Deployment"
+    finally:
+        Provider._models = original_models

webui/src/locales/en-US/model.json

@@ -119,7 +119,16 @@
     "loadFailed": "Failed to load provider catalog",
     "noModelsToTest": "No enabled models to test",
     "batchTestDone": "Batch test complete",
-    "batchTestSummary": "{{success}} succeeded, {{failed}} failed"
+    "batchTestSummary": "{{success}} succeeded, {{failed}} failed",
+    "azureDeploymentName": "Azure Deployment Name",
+    "azureDeploymentPlaceholder": "e.g. my-gpt-4o-prod",
+    "azureDeploymentHint": "Azure OpenAI requests use the deployment name, not a fixed model name. The preset models are examples; enter your own deployment name here.",
+    "azureDeploymentDisplayName": "Display Name (optional)",
+    "azureDeploymentDisplayPlaceholder": "e.g. GPT-4o Production",
+    "azureDeploymentRequired": "Select at least one preset model or enter an Azure deployment name",
+    "azureModelIdHint": "For Azure OpenAI, Model ID should be the deployment name from Azure Portal.",
+    "azureCustomDeployments": "Custom Azure Deployments",
+    "azureNoCustomDeployments": "No custom Azure deployment has been added yet."
   },
   "wizard": {
     "providerSaved": "Provider Saved",

webui/src/locales/zh-CN/model.json

@@ -119,7 +119,16 @@
     "loadFailed": "加载 Provider 目录失败",
     "noModelsToTest": "没有已启用的模型可测试",
     "batchTestDone": "批量测试完成",
-    "batchTestSummary": "{{success}} 成功, {{failed}} 失败"
+    "batchTestSummary": "{{success}} 成功, {{failed}} 失败",
+    "azureDeploymentName": "Azure 部署名称",
+    "azureDeploymentPlaceholder": "例如 my-gpt-4o-prod",
+    "azureDeploymentHint": "Azure OpenAI 请求使用 deployment name，而不是固定模型名。预设模型只是常用示例，你可以在这里填写自己的部署名称。",
+    "azureDeploymentDisplayName": "显示名称（可选）",
+    "azureDeploymentDisplayPlaceholder": "例如 GPT-4o Production",
+    "azureDeploymentRequired": "请至少选择一个预设模型，或填写 Azure deployment name",
+    "azureModelIdHint": "对于 Azure OpenAI，模型 ID 请填写 Azure Portal 中的 deployment name。",
+    "azureCustomDeployments": "自定义 Azure Deployments",
+    "azureNoCustomDeployments": "尚未添加自定义 Azure deployment。"
   },
   "wizard": {
     "providerSaved": "Provider 已保存",